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 #include <linux/uidgid.h>
41 #include <linux/crypto.h>
42 #include <linux/key.h>
44 #include <libcfs/libcfs.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
48 #include <lustre_net.h>
49 #include <lustre_import.h>
50 #include <lustre_dlm.h>
51 #include <lustre_sec.h>
53 #include "ptlrpc_internal.h"
55 static int send_sepol;
56 module_param(send_sepol, int, 0644);
57 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
63 static rwlock_t policy_lock;
64 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
68 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
70 __u16 number = policy->sp_policy;
72 LASSERT(policy->sp_name);
73 LASSERT(policy->sp_cops);
74 LASSERT(policy->sp_sops);
76 if (number >= SPTLRPC_POLICY_MAX)
79 write_lock(&policy_lock);
80 if (unlikely(policies[number])) {
81 write_unlock(&policy_lock);
84 policies[number] = policy;
85 write_unlock(&policy_lock);
87 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
90 EXPORT_SYMBOL(sptlrpc_register_policy);
92 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
94 __u16 number = policy->sp_policy;
96 LASSERT(number < SPTLRPC_POLICY_MAX);
98 write_lock(&policy_lock);
99 if (unlikely(policies[number] == NULL)) {
100 write_unlock(&policy_lock);
101 CERROR("%s: already unregistered\n", policy->sp_name);
105 LASSERT(policies[number] == policy);
106 policies[number] = NULL;
107 write_unlock(&policy_lock);
109 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
112 EXPORT_SYMBOL(sptlrpc_unregister_policy);
115 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
117 static DEFINE_MUTEX(load_mutex);
118 static atomic_t loaded = ATOMIC_INIT(0);
119 struct ptlrpc_sec_policy *policy;
120 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
123 if (number >= SPTLRPC_POLICY_MAX)
127 read_lock(&policy_lock);
128 policy = policies[number];
129 if (policy && !try_module_get(policy->sp_owner))
132 flag = atomic_read(&loaded);
133 read_unlock(&policy_lock);
135 if (policy != NULL || flag != 0 ||
136 number != SPTLRPC_POLICY_GSS)
139 /* try to load gss module, once */
140 mutex_lock(&load_mutex);
141 if (atomic_read(&loaded) == 0) {
142 if (request_module("ptlrpc_gss") == 0)
144 "module ptlrpc_gss loaded on demand\n");
146 CERROR("Unable to load module ptlrpc_gss\n");
148 atomic_set(&loaded, 1);
150 mutex_unlock(&load_mutex);
156 __u32 sptlrpc_name2flavor_base(const char *name)
158 if (!strcmp(name, "null"))
159 return SPTLRPC_FLVR_NULL;
160 if (!strcmp(name, "plain"))
161 return SPTLRPC_FLVR_PLAIN;
162 if (!strcmp(name, "gssnull"))
163 return SPTLRPC_FLVR_GSSNULL;
164 if (!strcmp(name, "krb5n"))
165 return SPTLRPC_FLVR_KRB5N;
166 if (!strcmp(name, "krb5a"))
167 return SPTLRPC_FLVR_KRB5A;
168 if (!strcmp(name, "krb5i"))
169 return SPTLRPC_FLVR_KRB5I;
170 if (!strcmp(name, "krb5p"))
171 return SPTLRPC_FLVR_KRB5P;
172 if (!strcmp(name, "skn"))
173 return SPTLRPC_FLVR_SKN;
174 if (!strcmp(name, "ska"))
175 return SPTLRPC_FLVR_SKA;
176 if (!strcmp(name, "ski"))
177 return SPTLRPC_FLVR_SKI;
178 if (!strcmp(name, "skpi"))
179 return SPTLRPC_FLVR_SKPI;
181 return SPTLRPC_FLVR_INVALID;
183 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
185 const char *sptlrpc_flavor2name_base(__u32 flvr)
187 __u32 base = SPTLRPC_FLVR_BASE(flvr);
189 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
191 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
199 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
201 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
203 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
205 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
207 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
209 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
212 CERROR("invalid wire flavor 0x%x\n", flvr);
215 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
217 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
218 char *buf, int bufsize)
220 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
221 snprintf(buf, bufsize, "hash:%s",
222 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
224 snprintf(buf, bufsize, "%s",
225 sptlrpc_flavor2name_base(sf->sf_rpc));
227 buf[bufsize - 1] = '\0';
230 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
232 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
234 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
237 * currently we don't support customized bulk specification for
238 * flavors other than plain
240 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
244 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
245 strncat(buf, bspec, bufsize);
248 buf[bufsize - 1] = '\0';
251 EXPORT_SYMBOL(sptlrpc_flavor2name);
253 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
257 if (flags & PTLRPC_SEC_FL_REVERSE)
258 strlcat(buf, "reverse,", bufsize);
259 if (flags & PTLRPC_SEC_FL_ROOTONLY)
260 strlcat(buf, "rootonly,", bufsize);
261 if (flags & PTLRPC_SEC_FL_UDESC)
262 strlcat(buf, "udesc,", bufsize);
263 if (flags & PTLRPC_SEC_FL_BULK)
264 strlcat(buf, "bulk,", bufsize);
266 strlcat(buf, "-,", bufsize);
270 EXPORT_SYMBOL(sptlrpc_secflags2str);
273 * client context APIs
277 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
279 struct vfs_cred vcred;
280 int create = 1, remove_dead = 1;
283 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
285 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
286 PTLRPC_SEC_FL_ROOTONLY)) {
289 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
294 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
295 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
298 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
302 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
304 atomic_inc(&ctx->cc_refcount);
307 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
309 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
311 struct ptlrpc_sec *sec = ctx->cc_sec;
314 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
316 if (!atomic_dec_and_test(&ctx->cc_refcount))
319 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
321 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
324 * Expire the client context immediately.
326 * \pre Caller must hold at least 1 reference on the \a ctx.
328 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
330 LASSERT(ctx->cc_ops->die);
331 ctx->cc_ops->die(ctx, 0);
333 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
336 * To wake up the threads who are waiting for this client context. Called
337 * after some status change happened on \a ctx.
339 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
341 struct ptlrpc_request *req, *next;
343 spin_lock(&ctx->cc_lock);
344 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
346 list_del_init(&req->rq_ctx_chain);
347 ptlrpc_client_wake_req(req);
349 spin_unlock(&ctx->cc_lock);
351 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
353 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
355 LASSERT(ctx->cc_ops);
357 if (ctx->cc_ops->display == NULL)
360 return ctx->cc_ops->display(ctx, buf, bufsize);
363 static int import_sec_check_expire(struct obd_import *imp)
367 write_lock(&imp->imp_sec_lock);
368 if (imp->imp_sec_expire &&
369 imp->imp_sec_expire < ktime_get_real_seconds()) {
371 imp->imp_sec_expire = 0;
373 write_unlock(&imp->imp_sec_lock);
378 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
379 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
383 * Get and validate the client side ptlrpc security facilities from
384 * \a imp. There is a race condition on client reconnect when the import is
385 * being destroyed while there are outstanding client bound requests. In
386 * this case do not output any error messages if import secuity is not
389 * \param[in] imp obd import associated with client
390 * \param[out] sec client side ptlrpc security
392 * \retval 0 if security retrieved successfully
393 * \retval -ve errno if there was a problem
395 static int import_sec_validate_get(struct obd_import *imp,
396 struct ptlrpc_sec **sec)
400 if (unlikely(imp->imp_sec_expire)) {
401 rc = import_sec_check_expire(imp);
406 *sec = sptlrpc_import_sec_ref(imp);
408 CERROR("import %p (%s) with no sec\n",
409 imp, ptlrpc_import_state_name(imp->imp_state));
413 if (unlikely((*sec)->ps_dying)) {
414 CERROR("attempt to use dying sec %p\n", sec);
415 sptlrpc_sec_put(*sec);
423 * Given a \a req, find or allocate an appropriate context for it.
424 * \pre req->rq_cli_ctx == NULL.
426 * \retval 0 succeed, and req->rq_cli_ctx is set.
427 * \retval -ev error number, and req->rq_cli_ctx == NULL.
429 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
431 struct obd_import *imp = req->rq_import;
432 struct ptlrpc_sec *sec;
437 LASSERT(!req->rq_cli_ctx);
440 rc = import_sec_validate_get(imp, &sec);
444 req->rq_cli_ctx = get_my_ctx(sec);
446 sptlrpc_sec_put(sec);
448 if (!req->rq_cli_ctx) {
449 CERROR("req %p: fail to get context\n", req);
450 RETURN(-ECONNREFUSED);
457 * Drop the context for \a req.
458 * \pre req->rq_cli_ctx != NULL.
459 * \post req->rq_cli_ctx == NULL.
461 * If \a sync == 0, this function should return quickly without sleep;
462 * otherwise it might trigger and wait for the whole process of sending
463 * an context-destroying rpc to server.
465 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
470 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);
502 "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
503 req, oldctx, oldctx->cc_vcred.vc_uid,
504 sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
505 sec2target_str(newctx->cc_sec), oldctx->cc_sec,
506 oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
507 newctx->cc_sec->ps_policy->sp_name);
510 old_flvr = req->rq_flvr;
512 /* save request message */
513 reqmsg_size = req->rq_reqlen;
514 if (reqmsg_size != 0) {
515 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
518 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
521 /* release old req/rep buf */
522 req->rq_cli_ctx = oldctx;
523 sptlrpc_cli_free_reqbuf(req);
524 sptlrpc_cli_free_repbuf(req);
525 req->rq_cli_ctx = newctx;
527 /* recalculate the flavor */
528 sptlrpc_req_set_flavor(req, 0);
531 * alloc new request buffer
532 * we don't need to alloc reply buffer here, leave it to the
533 * rest procedure of ptlrpc
535 if (reqmsg_size != 0) {
536 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
538 LASSERT(req->rq_reqmsg);
539 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
541 CWARN("failed to alloc reqbuf: %d\n", rc);
542 req->rq_flvr = old_flvr;
545 OBD_FREE_LARGE(reqmsg, reqmsg_size);
551 * If current context of \a req is dead somehow, e.g. we just switched flavor
552 * thus marked original contexts dead, we'll find a new context for it. if
553 * no switch is needed, \a req will end up with the same context.
555 * \note a request must have a context, to keep other parts of code happy.
556 * In any case of failure during the switching, we must restore the old one.
558 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
560 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
561 struct ptlrpc_cli_ctx *newctx;
568 sptlrpc_cli_ctx_get(oldctx);
569 sptlrpc_req_put_ctx(req, 0);
571 rc = sptlrpc_req_get_ctx(req);
573 LASSERT(!req->rq_cli_ctx);
575 /* restore old ctx */
576 req->rq_cli_ctx = oldctx;
580 newctx = req->rq_cli_ctx;
583 if (unlikely(newctx == oldctx &&
584 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
586 * still get the old dead ctx, usually means system too busy
589 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
590 newctx, newctx->cc_flags);
592 schedule_timeout_interruptible(cfs_time_seconds(1));
593 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
596 * new ctx not up to date yet
599 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
600 newctx, newctx->cc_flags);
603 * it's possible newctx == oldctx if we're switching
604 * subflavor with the same sec.
606 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
608 /* restore old ctx */
609 sptlrpc_req_put_ctx(req, 0);
610 req->rq_cli_ctx = oldctx;
614 LASSERT(req->rq_cli_ctx == newctx);
617 sptlrpc_cli_ctx_put(oldctx, 1);
620 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
623 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
625 if (cli_ctx_is_refreshed(ctx))
631 void ctx_refresh_interrupt(struct ptlrpc_request *req)
634 spin_lock(&req->rq_lock);
636 spin_unlock(&req->rq_lock);
640 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
642 spin_lock(&ctx->cc_lock);
643 if (!list_empty(&req->rq_ctx_chain))
644 list_del_init(&req->rq_ctx_chain);
645 spin_unlock(&ctx->cc_lock);
649 * To refresh the context of \req, if it's not up-to-date.
651 * - == 0: do not wait
652 * - == MAX_SCHEDULE_TIMEOUT: wait indefinitely
653 * - > 0: not supported
655 * The status of the context could be subject to be changed by other threads
656 * at any time. We allow this race, but once we return with 0, the caller will
657 * suppose it's uptodated and keep using it until the owning rpc is done.
659 * \retval 0 only if the context is uptodated.
660 * \retval -ev error number.
662 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
664 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
665 struct ptlrpc_sec *sec;
672 if (req->rq_ctx_init || req->rq_ctx_fini)
675 if (timeout != 0 && timeout != MAX_SCHEDULE_TIMEOUT) {
676 CERROR("req %p: invalid timeout %lu\n", req, timeout);
681 * during the process a request's context might change type even
682 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
686 rc = import_sec_validate_get(req->rq_import, &sec);
690 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
691 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
692 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
693 req_off_ctx_list(req, ctx);
694 sptlrpc_req_replace_dead_ctx(req);
695 ctx = req->rq_cli_ctx;
697 sptlrpc_sec_put(sec);
699 if (cli_ctx_is_eternal(ctx))
702 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
703 if (ctx->cc_ops->refresh)
704 ctx->cc_ops->refresh(ctx);
706 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
708 LASSERT(ctx->cc_ops->validate);
709 if (ctx->cc_ops->validate(ctx) == 0) {
710 req_off_ctx_list(req, ctx);
714 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
715 spin_lock(&req->rq_lock);
717 spin_unlock(&req->rq_lock);
718 req_off_ctx_list(req, ctx);
723 * There's a subtle issue for resending RPCs, suppose following
725 * 1. the request was sent to server.
726 * 2. recovery was kicked start, after finished the request was
728 * 3. resend the request.
729 * 4. old reply from server received, we accept and verify the reply.
730 * this has to be success, otherwise the error will be aware
732 * 5. new reply from server received, dropped by LNet.
734 * Note the xid of old & new request is the same. We can't simply
735 * change xid for the resent request because the server replies on
736 * it for reply reconstruction.
738 * Commonly the original context should be uptodate because we
739 * have an expiry nice time; server will keep its context because
740 * we at least hold a ref of old context which prevent context
741 * from destroying RPC being sent. So server still can accept the
742 * request and finish the RPC. But if that's not the case:
743 * 1. If server side context has been trimmed, a NO_CONTEXT will
744 * be returned, gss_cli_ctx_verify/unseal will switch to new
746 * 2. Current context never be refreshed, then we are fine: we
747 * never really send request with old context before.
749 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
750 unlikely(req->rq_reqmsg) &&
751 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
752 req_off_ctx_list(req, ctx);
756 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
757 req_off_ctx_list(req, ctx);
759 * don't switch ctx if import was deactivated
761 if (req->rq_import->imp_deactive) {
762 spin_lock(&req->rq_lock);
764 spin_unlock(&req->rq_lock);
768 rc = sptlrpc_req_replace_dead_ctx(req);
770 LASSERT(ctx == req->rq_cli_ctx);
771 CERROR("req %p: failed to replace dead ctx %p: %d\n",
773 spin_lock(&req->rq_lock);
775 spin_unlock(&req->rq_lock);
779 ctx = req->rq_cli_ctx;
784 * Now we're sure this context is during upcall, add myself into
787 spin_lock(&ctx->cc_lock);
788 if (list_empty(&req->rq_ctx_chain))
789 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
790 spin_unlock(&ctx->cc_lock);
793 RETURN(-EWOULDBLOCK);
795 /* Clear any flags that may be present from previous sends */
796 LASSERT(req->rq_receiving_reply == 0);
797 spin_lock(&req->rq_lock);
799 req->rq_timedout = 0;
802 spin_unlock(&req->rq_lock);
804 /* by now we know that timeout value is MAX_SCHEDULE_TIMEOUT,
805 * so wait indefinitely with non-fatal signals blocked
807 if (l_wait_event_abortable(req->rq_reply_waitq,
808 ctx_check_refresh(ctx)) == -ERESTARTSYS) {
810 ctx_refresh_interrupt(req);
814 * following cases could lead us here:
815 * - successfully refreshed;
817 * - timedout, and we don't want recover from the failure;
818 * - timedout, and waked up upon recovery finished;
819 * - someone else mark this ctx dead by force;
820 * - someone invalidate the req and call ptlrpc_client_wake_req(),
821 * e.g. ptlrpc_abort_inflight();
823 if (!cli_ctx_is_refreshed(ctx)) {
824 /* timed out or interruptted */
825 req_off_ctx_list(req, ctx);
834 /* Bring ptlrpc_sec context up-to-date */
835 int sptlrpc_export_update_ctx(struct obd_export *exp)
837 struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
838 struct ptlrpc_sec *sec = NULL;
839 struct ptlrpc_cli_ctx *ctx = NULL;
843 sec = sptlrpc_import_sec_ref(imp);
845 ctx = get_my_ctx(sec);
846 sptlrpc_sec_put(sec);
850 if (ctx->cc_ops->refresh)
851 rc = ctx->cc_ops->refresh(ctx);
852 sptlrpc_cli_ctx_put(ctx, 1);
858 * Initialize flavor settings for \a req, according to \a opcode.
860 * \note this could be called in two situations:
861 * - new request from ptlrpc_pre_req(), with proper @opcode
862 * - old request which changed ctx in the middle, with @opcode == 0
864 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
866 struct ptlrpc_sec *sec;
868 LASSERT(req->rq_import);
869 LASSERT(req->rq_cli_ctx);
870 LASSERT(req->rq_cli_ctx->cc_sec);
871 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
873 /* special security flags according to opcode */
877 case MGS_CONFIG_READ:
879 req->rq_bulk_read = 1;
883 req->rq_bulk_write = 1;
886 req->rq_ctx_init = 1;
889 req->rq_ctx_fini = 1;
892 /* init/fini rpc won't be resend, so can't be here */
893 LASSERT(req->rq_ctx_init == 0);
894 LASSERT(req->rq_ctx_fini == 0);
896 /* cleanup flags, which should be recalculated */
897 req->rq_pack_udesc = 0;
898 req->rq_pack_bulk = 0;
902 sec = req->rq_cli_ctx->cc_sec;
904 spin_lock(&sec->ps_lock);
905 req->rq_flvr = sec->ps_flvr;
906 spin_unlock(&sec->ps_lock);
909 * force SVC_NULL for context initiation rpc, SVC_INTG for context
912 if (unlikely(req->rq_ctx_init))
913 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
914 else if (unlikely(req->rq_ctx_fini))
915 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
917 /* user descriptor flag, null security can't do it anyway */
918 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
919 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
920 req->rq_pack_udesc = 1;
922 /* bulk security flag */
923 if ((req->rq_bulk_read || req->rq_bulk_write) &&
924 sptlrpc_flavor_has_bulk(&req->rq_flvr))
925 req->rq_pack_bulk = 1;
928 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
930 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
933 LASSERT(req->rq_clrbuf);
934 if (req->rq_pool || !req->rq_reqbuf)
937 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
938 req->rq_reqbuf = NULL;
939 req->rq_reqbuf_len = 0;
943 * Given an import \a imp, check whether current user has a valid context
944 * or not. We may create a new context and try to refresh it, and try
945 * repeatedly try in case of non-fatal errors. Return 0 means success.
947 int sptlrpc_import_check_ctx(struct obd_import *imp)
949 struct ptlrpc_sec *sec;
950 struct ptlrpc_cli_ctx *ctx;
951 struct ptlrpc_request *req = NULL;
958 sec = sptlrpc_import_sec_ref(imp);
959 ctx = get_my_ctx(sec);
960 sptlrpc_sec_put(sec);
965 if (cli_ctx_is_eternal(ctx) ||
966 ctx->cc_ops->validate(ctx) == 0) {
967 sptlrpc_cli_ctx_put(ctx, 1);
971 if (cli_ctx_is_error(ctx)) {
972 sptlrpc_cli_ctx_put(ctx, 1);
976 req = ptlrpc_request_cache_alloc(GFP_NOFS);
980 ptlrpc_cli_req_init(req);
981 atomic_set(&req->rq_refcount, 10000);
983 req->rq_import = imp;
984 req->rq_flvr = sec->ps_flvr;
985 req->rq_cli_ctx = ctx;
987 rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
988 LASSERT(list_empty(&req->rq_ctx_chain));
989 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
990 ptlrpc_request_cache_free(req);
996 * Used by ptlrpc client, to perform the pre-defined security transformation
997 * upon the request message of \a req. After this function called,
998 * req->rq_reqmsg is still accessible as clear text.
1000 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1002 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1008 LASSERT(ctx->cc_sec);
1009 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1012 * we wrap bulk request here because now we can be sure
1013 * the context is uptodate.
1016 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1021 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1022 case SPTLRPC_SVC_NULL:
1023 case SPTLRPC_SVC_AUTH:
1024 case SPTLRPC_SVC_INTG:
1025 LASSERT(ctx->cc_ops->sign);
1026 rc = ctx->cc_ops->sign(ctx, req);
1028 case SPTLRPC_SVC_PRIV:
1029 LASSERT(ctx->cc_ops->seal);
1030 rc = ctx->cc_ops->seal(ctx, req);
1037 LASSERT(req->rq_reqdata_len);
1038 LASSERT(req->rq_reqdata_len % 8 == 0);
1039 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1045 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1047 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1053 LASSERT(ctx->cc_sec);
1054 LASSERT(req->rq_repbuf);
1055 LASSERT(req->rq_repdata);
1056 LASSERT(req->rq_repmsg == NULL);
1058 req->rq_rep_swab_mask = 0;
1060 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1063 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1067 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1071 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1072 CERROR("replied data length %d too small\n",
1073 req->rq_repdata_len);
1077 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1078 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1079 CERROR("reply policy %u doesn't match request policy %u\n",
1080 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1081 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1085 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1086 case SPTLRPC_SVC_NULL:
1087 case SPTLRPC_SVC_AUTH:
1088 case SPTLRPC_SVC_INTG:
1089 LASSERT(ctx->cc_ops->verify);
1090 rc = ctx->cc_ops->verify(ctx, req);
1092 case SPTLRPC_SVC_PRIV:
1093 LASSERT(ctx->cc_ops->unseal);
1094 rc = ctx->cc_ops->unseal(ctx, req);
1099 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1101 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1103 req->rq_rep_swab_mask = 0;
1108 * Used by ptlrpc client, to perform security transformation upon the reply
1109 * message of \a req. After return successfully, req->rq_repmsg points to
1110 * the reply message in clear text.
1112 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1115 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1117 LASSERT(req->rq_repbuf);
1118 LASSERT(req->rq_repdata == NULL);
1119 LASSERT(req->rq_repmsg == NULL);
1120 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1122 if (req->rq_reply_off == 0 &&
1123 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1124 CERROR("real reply with offset 0\n");
1128 if (req->rq_reply_off % 8 != 0) {
1129 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1133 req->rq_repdata = (struct lustre_msg *)
1134 (req->rq_repbuf + req->rq_reply_off);
1135 req->rq_repdata_len = req->rq_nob_received;
1137 return do_cli_unwrap_reply(req);
1141 * Used by ptlrpc client, to perform security transformation upon the early
1142 * reply message of \a req. We expect the rq_reply_off is 0, and
1143 * rq_nob_received is the early reply size.
1145 * Because the receive buffer might be still posted, the reply data might be
1146 * changed at any time, no matter we're holding rq_lock or not. For this reason
1147 * we allocate a separate ptlrpc_request and reply buffer for early reply
1150 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1151 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1152 * \a *req_ret to release it.
1153 * \retval -ev error number, and \a req_ret will not be set.
1155 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1156 struct ptlrpc_request **req_ret)
1158 struct ptlrpc_request *early_req;
1160 int early_bufsz, early_size;
1165 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1166 if (early_req == NULL)
1169 ptlrpc_cli_req_init(early_req);
1171 early_size = req->rq_nob_received;
1172 early_bufsz = size_roundup_power2(early_size);
1173 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1174 if (early_buf == NULL)
1175 GOTO(err_req, rc = -ENOMEM);
1177 /* sanity checkings and copy data out, do it inside spinlock */
1178 spin_lock(&req->rq_lock);
1180 if (req->rq_replied) {
1181 spin_unlock(&req->rq_lock);
1182 GOTO(err_buf, rc = -EALREADY);
1185 LASSERT(req->rq_repbuf);
1186 LASSERT(req->rq_repdata == NULL);
1187 LASSERT(req->rq_repmsg == NULL);
1189 if (req->rq_reply_off != 0) {
1190 CERROR("early reply with offset %u\n", req->rq_reply_off);
1191 spin_unlock(&req->rq_lock);
1192 GOTO(err_buf, rc = -EPROTO);
1195 if (req->rq_nob_received != early_size) {
1196 /* even another early arrived the size should be the same */
1197 CERROR("data size has changed from %u to %u\n",
1198 early_size, req->rq_nob_received);
1199 spin_unlock(&req->rq_lock);
1200 GOTO(err_buf, rc = -EINVAL);
1203 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1204 CERROR("early reply length %d too small\n",
1205 req->rq_nob_received);
1206 spin_unlock(&req->rq_lock);
1207 GOTO(err_buf, rc = -EALREADY);
1210 memcpy(early_buf, req->rq_repbuf, early_size);
1211 spin_unlock(&req->rq_lock);
1213 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1214 early_req->rq_flvr = req->rq_flvr;
1215 early_req->rq_repbuf = early_buf;
1216 early_req->rq_repbuf_len = early_bufsz;
1217 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1218 early_req->rq_repdata_len = early_size;
1219 early_req->rq_early = 1;
1220 early_req->rq_reqmsg = req->rq_reqmsg;
1222 rc = do_cli_unwrap_reply(early_req);
1224 DEBUG_REQ(D_ADAPTTO, early_req,
1225 "unwrap early reply: rc = %d", rc);
1229 LASSERT(early_req->rq_repmsg);
1230 *req_ret = early_req;
1234 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1236 OBD_FREE_LARGE(early_buf, early_bufsz);
1238 ptlrpc_request_cache_free(early_req);
1243 * Used by ptlrpc client, to release a processed early reply \a early_req.
1245 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1247 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1249 LASSERT(early_req->rq_repbuf);
1250 LASSERT(early_req->rq_repdata);
1251 LASSERT(early_req->rq_repmsg);
1253 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1254 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1255 ptlrpc_request_cache_free(early_req);
1258 /**************************************************
1260 **************************************************/
1263 * "fixed" sec (e.g. null) use sec_id < 0
1265 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1267 int sptlrpc_get_next_secid(void)
1269 return atomic_inc_return(&sptlrpc_sec_id);
1271 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1274 * client side high-level security APIs
1277 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1278 int grace, int force)
1280 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1282 LASSERT(policy->sp_cops);
1283 LASSERT(policy->sp_cops->flush_ctx_cache);
1285 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1288 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1290 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1292 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1293 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1294 LASSERT(policy->sp_cops->destroy_sec);
1296 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1298 policy->sp_cops->destroy_sec(sec);
1299 sptlrpc_policy_put(policy);
1302 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1304 sec_cop_destroy_sec(sec);
1306 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1308 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1310 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1312 if (sec->ps_policy->sp_cops->kill_sec) {
1313 sec->ps_policy->sp_cops->kill_sec(sec);
1315 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1319 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1322 atomic_inc(&sec->ps_refcount);
1326 EXPORT_SYMBOL(sptlrpc_sec_get);
1328 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1331 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1333 if (atomic_dec_and_test(&sec->ps_refcount)) {
1334 sptlrpc_gc_del_sec(sec);
1335 sec_cop_destroy_sec(sec);
1339 EXPORT_SYMBOL(sptlrpc_sec_put);
1342 * policy module is responsible for taking refrence of import
1345 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1346 struct ptlrpc_svc_ctx *svc_ctx,
1347 struct sptlrpc_flavor *sf,
1348 enum lustre_sec_part sp)
1350 struct ptlrpc_sec_policy *policy;
1351 struct ptlrpc_sec *sec;
1357 LASSERT(imp->imp_dlm_fake == 1);
1359 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1360 imp->imp_obd->obd_type->typ_name,
1361 imp->imp_obd->obd_name,
1362 sptlrpc_flavor2name(sf, str, sizeof(str)));
1364 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1365 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1367 LASSERT(imp->imp_dlm_fake == 0);
1369 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1370 imp->imp_obd->obd_type->typ_name,
1371 imp->imp_obd->obd_name,
1372 sptlrpc_flavor2name(sf, str, sizeof(str)));
1374 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1376 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1381 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1383 atomic_inc(&sec->ps_refcount);
1387 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1388 sptlrpc_gc_add_sec(sec);
1390 sptlrpc_policy_put(policy);
1396 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1398 struct ptlrpc_sec *sec;
1400 read_lock(&imp->imp_sec_lock);
1401 sec = sptlrpc_sec_get(imp->imp_sec);
1402 read_unlock(&imp->imp_sec_lock);
1406 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1408 static void sptlrpc_import_sec_install(struct obd_import *imp,
1409 struct ptlrpc_sec *sec)
1411 struct ptlrpc_sec *old_sec;
1413 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1415 write_lock(&imp->imp_sec_lock);
1416 old_sec = imp->imp_sec;
1418 write_unlock(&imp->imp_sec_lock);
1421 sptlrpc_sec_kill(old_sec);
1423 /* balance the ref taken by this import */
1424 sptlrpc_sec_put(old_sec);
1429 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1431 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1435 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1441 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1442 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1444 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1445 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1447 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1448 struct ptlrpc_svc_ctx *svc_ctx,
1449 struct sptlrpc_flavor *flvr)
1451 struct ptlrpc_connection *conn;
1452 struct sptlrpc_flavor sf;
1453 struct ptlrpc_sec *sec, *newsec;
1454 enum lustre_sec_part sp;
1465 conn = imp->imp_connection;
1467 if (svc_ctx == NULL) {
1468 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1470 * normal import, determine flavor from rule set, except
1471 * for mgc the flavor is predetermined.
1473 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1474 sf = cliobd->cl_flvr_mgc;
1476 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1478 &cliobd->cl_target_uuid,
1481 sp = imp->imp_obd->u.cli.cl_sp_me;
1483 /* reverse import, determine flavor from incoming reqeust */
1486 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1487 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1488 PTLRPC_SEC_FL_ROOTONLY;
1490 sp = sptlrpc_target_sec_part(imp->imp_obd);
1493 sec = sptlrpc_import_sec_ref(imp);
1497 if (flavor_equal(&sf, &sec->ps_flvr))
1500 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1501 imp->imp_obd->obd_name,
1502 obd_uuid2str(&conn->c_remote_uuid),
1503 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1504 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1505 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1506 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1507 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1508 imp->imp_obd->obd_name,
1509 obd_uuid2str(&conn->c_remote_uuid),
1510 LNET_NIDNET(conn->c_self),
1511 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1514 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1516 sptlrpc_import_sec_install(imp, newsec);
1518 CERROR("import %s->%s: failed to create new sec\n",
1519 imp->imp_obd->obd_name,
1520 obd_uuid2str(&conn->c_remote_uuid));
1525 sptlrpc_sec_put(sec);
1529 void sptlrpc_import_sec_put(struct obd_import *imp)
1532 sptlrpc_sec_kill(imp->imp_sec);
1534 sptlrpc_sec_put(imp->imp_sec);
1535 imp->imp_sec = NULL;
1539 static void import_flush_ctx_common(struct obd_import *imp,
1540 uid_t uid, int grace, int force)
1542 struct ptlrpc_sec *sec;
1547 sec = sptlrpc_import_sec_ref(imp);
1551 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1552 sptlrpc_sec_put(sec);
1555 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1558 * it's important to use grace mode, see explain in
1559 * sptlrpc_req_refresh_ctx()
1561 import_flush_ctx_common(imp, 0, 1, 1);
1564 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1566 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1569 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1571 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1573 import_flush_ctx_common(imp, -1, 1, 1);
1575 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1578 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1579 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1581 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1583 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1584 struct ptlrpc_sec_policy *policy;
1588 LASSERT(ctx->cc_sec);
1589 LASSERT(ctx->cc_sec->ps_policy);
1590 LASSERT(req->rq_reqmsg == NULL);
1591 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1593 policy = ctx->cc_sec->ps_policy;
1594 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1596 LASSERT(req->rq_reqmsg);
1597 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1599 /* zeroing preallocated buffer */
1601 memset(req->rq_reqmsg, 0, msgsize);
1608 * Used by ptlrpc client to free request buffer of \a req. After this
1609 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1611 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1613 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1614 struct ptlrpc_sec_policy *policy;
1617 LASSERT(ctx->cc_sec);
1618 LASSERT(ctx->cc_sec->ps_policy);
1619 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1621 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1624 policy = ctx->cc_sec->ps_policy;
1625 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1626 req->rq_reqmsg = NULL;
1630 * NOTE caller must guarantee the buffer size is enough for the enlargement
1632 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1633 int segment, int newsize)
1636 int oldsize, oldmsg_size, movesize;
1638 LASSERT(segment < msg->lm_bufcount);
1639 LASSERT(msg->lm_buflens[segment] <= newsize);
1641 if (msg->lm_buflens[segment] == newsize)
1644 /* nothing to do if we are enlarging the last segment */
1645 if (segment == msg->lm_bufcount - 1) {
1646 msg->lm_buflens[segment] = newsize;
1650 oldsize = msg->lm_buflens[segment];
1652 src = lustre_msg_buf(msg, segment + 1, 0);
1653 msg->lm_buflens[segment] = newsize;
1654 dst = lustre_msg_buf(msg, segment + 1, 0);
1655 msg->lm_buflens[segment] = oldsize;
1657 /* move from segment + 1 to end segment */
1658 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1659 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1660 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1661 LASSERT(movesize >= 0);
1664 memmove(dst, src, movesize);
1666 /* note we don't clear the ares where old data live, not secret */
1668 /* finally set new segment size */
1669 msg->lm_buflens[segment] = newsize;
1671 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1674 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1675 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1676 * preserved after the enlargement. this must be called after original request
1677 * buffer being allocated.
1679 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1680 * so caller should refresh its local pointers if needed.
1682 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1683 const struct req_msg_field *field,
1686 struct req_capsule *pill = &req->rq_pill;
1687 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1688 struct ptlrpc_sec_cops *cops;
1689 struct lustre_msg *msg = req->rq_reqmsg;
1690 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1694 LASSERT(msg->lm_bufcount > segment);
1695 LASSERT(msg->lm_buflens[segment] <= newsize);
1697 if (msg->lm_buflens[segment] == newsize)
1700 cops = ctx->cc_sec->ps_policy->sp_cops;
1701 LASSERT(cops->enlarge_reqbuf);
1702 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1704 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1707 * Used by ptlrpc client to allocate reply buffer of \a req.
1709 * \note After this, req->rq_repmsg is still not accessible.
1711 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1713 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1714 struct ptlrpc_sec_policy *policy;
1719 LASSERT(ctx->cc_sec);
1720 LASSERT(ctx->cc_sec->ps_policy);
1725 policy = ctx->cc_sec->ps_policy;
1726 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1730 * Used by ptlrpc client to free reply buffer of \a req. After this
1731 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1733 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1735 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1736 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] = (char *)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 (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1808 imp->imp_sec->ps_sepol[0] == '\0') {
1809 /* ps_sepol has not been initialized */
1815 mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1816 snprintf(mtime_str, sizeof(mtime_str), "%lld",
1817 mtime_ms / MSEC_PER_SEC);
1818 mode_str[0] = imp->imp_sec->ps_sepol[0];
1820 spin_unlock(&imp->imp_sec->ps_lock);
1821 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1828 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1832 if (send_sepol == 0 || !selinux_is_enabled())
1835 if (send_sepol == -1)
1836 /* send_sepol == -1 means fetch sepol status every time */
1839 spin_lock(&imp_sec->ps_lock);
1840 checknext = imp_sec->ps_sepol_checknext;
1841 spin_unlock(&imp_sec->ps_lock);
1843 /* next check is too far in time, please update */
1844 if (ktime_after(checknext,
1845 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1848 if (ktime_before(ktime_get(), checknext))
1849 /* too early to fetch sepol status */
1853 /* define new sepol_checknext time */
1854 spin_lock(&imp_sec->ps_lock);
1855 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1856 ktime_set(send_sepol, 0));
1857 spin_unlock(&imp_sec->ps_lock);
1862 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1864 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1869 (req->rq_sepol)[0] = '\0';
1871 #ifndef HAVE_SELINUX
1872 if (unlikely(send_sepol != 0))
1874 "Client cannot report SELinux status, it was not built against libselinux.\n");
1878 if (send_sepol == 0 || !selinux_is_enabled())
1881 if (imp_sec == NULL)
1884 /* Retrieve SELinux status info */
1885 if (sptlrpc_sepol_needs_check(imp_sec))
1886 rc = sepol_helper(req->rq_import);
1887 if (likely(rc == 0)) {
1888 spin_lock(&imp_sec->ps_lock);
1889 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1890 sizeof(req->rq_sepol));
1891 spin_unlock(&imp_sec->ps_lock);
1896 EXPORT_SYMBOL(sptlrpc_get_sepol);
1899 * server side security
1902 static int flavor_allowed(struct sptlrpc_flavor *exp,
1903 struct ptlrpc_request *req)
1905 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1907 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1910 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1911 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1912 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1913 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1919 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1922 * Given an export \a exp, check whether the flavor of incoming \a req
1923 * is allowed by the export \a exp. Main logic is about taking care of
1924 * changing configurations. Return 0 means success.
1926 int sptlrpc_target_export_check(struct obd_export *exp,
1927 struct ptlrpc_request *req)
1929 struct sptlrpc_flavor flavor;
1935 * client side export has no imp_reverse, skip
1936 * FIXME maybe we should check flavor this as well???
1938 if (exp->exp_imp_reverse == NULL)
1941 /* don't care about ctx fini rpc */
1942 if (req->rq_ctx_fini)
1945 spin_lock(&exp->exp_lock);
1948 * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1949 * the first req with the new flavor, then treat it as current flavor,
1950 * adapt reverse sec according to it.
1951 * note the first rpc with new flavor might not be with root ctx, in
1952 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1954 if (unlikely(exp->exp_flvr_changed) &&
1955 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1957 * make the new flavor as "current", and old ones as
1960 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1961 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1962 flavor = exp->exp_flvr_old[1];
1963 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1964 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1965 exp->exp_flvr_old[0] = exp->exp_flvr;
1966 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1967 EXP_FLVR_UPDATE_EXPIRE;
1968 exp->exp_flvr = flavor;
1970 /* flavor change finished */
1971 exp->exp_flvr_changed = 0;
1972 LASSERT(exp->exp_flvr_adapt == 1);
1974 /* if it's gss, we only interested in root ctx init */
1975 if (req->rq_auth_gss &&
1976 !(req->rq_ctx_init &&
1977 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1978 req->rq_auth_usr_ost))) {
1979 spin_unlock(&exp->exp_lock);
1980 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1981 req->rq_auth_gss, req->rq_ctx_init,
1982 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1983 req->rq_auth_usr_ost);
1987 exp->exp_flvr_adapt = 0;
1988 spin_unlock(&exp->exp_lock);
1990 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1991 req->rq_svc_ctx, &flavor);
1995 * if it equals to the current flavor, we accept it, but need to
1996 * dealing with reverse sec/ctx
1998 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
2000 * most cases should return here, we only interested in
2003 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2004 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2005 !req->rq_auth_usr_ost)) {
2006 spin_unlock(&exp->exp_lock);
2011 * if flavor just changed, we should not proceed, just leave
2012 * it and current flavor will be discovered and replaced
2013 * shortly, and let _this_ rpc pass through
2015 if (exp->exp_flvr_changed) {
2016 LASSERT(exp->exp_flvr_adapt);
2017 spin_unlock(&exp->exp_lock);
2021 if (exp->exp_flvr_adapt) {
2022 exp->exp_flvr_adapt = 0;
2023 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2024 exp, exp->exp_flvr.sf_rpc,
2025 exp->exp_flvr_old[0].sf_rpc,
2026 exp->exp_flvr_old[1].sf_rpc);
2027 flavor = exp->exp_flvr;
2028 spin_unlock(&exp->exp_lock);
2030 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2035 "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2036 exp, exp->exp_flvr.sf_rpc,
2037 exp->exp_flvr_old[0].sf_rpc,
2038 exp->exp_flvr_old[1].sf_rpc);
2039 spin_unlock(&exp->exp_lock);
2041 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2046 if (exp->exp_flvr_expire[0]) {
2047 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2048 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2050 "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2051 exp, exp->exp_flvr.sf_rpc,
2052 exp->exp_flvr_old[0].sf_rpc,
2053 exp->exp_flvr_old[1].sf_rpc,
2054 (s64)(exp->exp_flvr_expire[0] -
2055 ktime_get_real_seconds()));
2056 spin_unlock(&exp->exp_lock);
2060 CDEBUG(D_SEC, "mark middle expired\n");
2061 exp->exp_flvr_expire[0] = 0;
2063 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2064 exp->exp_flvr.sf_rpc,
2065 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2066 req->rq_flvr.sf_rpc);
2070 * now it doesn't match the current flavor, the only chance we can
2071 * accept it is match the old flavors which is not expired.
2073 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2074 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2075 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2076 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2078 exp->exp_flvr.sf_rpc,
2079 exp->exp_flvr_old[0].sf_rpc,
2080 exp->exp_flvr_old[1].sf_rpc,
2081 (s64)(exp->exp_flvr_expire[1] -
2082 ktime_get_real_seconds()));
2083 spin_unlock(&exp->exp_lock);
2087 CDEBUG(D_SEC, "mark oldest expired\n");
2088 exp->exp_flvr_expire[1] = 0;
2090 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2091 exp, exp->exp_flvr.sf_rpc,
2092 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2093 req->rq_flvr.sf_rpc);
2095 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2096 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2097 exp->exp_flvr_old[1].sf_rpc);
2100 spin_unlock(&exp->exp_lock);
2102 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2103 exp, exp->exp_obd->obd_name,
2104 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2105 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2106 req->rq_flvr.sf_rpc,
2107 exp->exp_flvr.sf_rpc,
2108 exp->exp_flvr_old[0].sf_rpc,
2109 exp->exp_flvr_expire[0] ?
2110 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2111 exp->exp_flvr_old[1].sf_rpc,
2112 exp->exp_flvr_expire[1] ?
2113 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2116 EXPORT_SYMBOL(sptlrpc_target_export_check);
2118 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2119 struct sptlrpc_rule_set *rset)
2121 struct obd_export *exp;
2122 struct sptlrpc_flavor new_flvr;
2126 spin_lock(&obd->obd_dev_lock);
2128 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2129 if (exp->exp_connection == NULL)
2133 * note if this export had just been updated flavor
2134 * (exp_flvr_changed == 1), this will override the
2137 spin_lock(&exp->exp_lock);
2138 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2139 exp->exp_connection->c_peer.nid,
2141 if (exp->exp_flvr_changed ||
2142 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2143 exp->exp_flvr_old[1] = new_flvr;
2144 exp->exp_flvr_expire[1] = 0;
2145 exp->exp_flvr_changed = 1;
2146 exp->exp_flvr_adapt = 1;
2148 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2149 exp, sptlrpc_part2name(exp->exp_sp_peer),
2150 exp->exp_flvr.sf_rpc,
2151 exp->exp_flvr_old[1].sf_rpc);
2153 spin_unlock(&exp->exp_lock);
2156 spin_unlock(&obd->obd_dev_lock);
2158 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2160 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2162 /* peer's claim is unreliable unless gss is being used */
2163 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2166 switch (req->rq_sp_from) {
2168 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2169 /* The below message is checked in sanity-sec test_33 */
2170 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2171 svc_rc = SECSVC_DROP;
2175 if (!req->rq_auth_usr_mdt) {
2176 /* The below message is checked in sanity-sec test_33 */
2177 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2178 svc_rc = SECSVC_DROP;
2182 if (!req->rq_auth_usr_ost) {
2183 /* The below message is checked in sanity-sec test_33 */
2184 DEBUG_REQ(D_ERROR, req, "faked source OST");
2185 svc_rc = SECSVC_DROP;
2190 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2191 !req->rq_auth_usr_ost) {
2192 /* The below message is checked in sanity-sec test_33 */
2193 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2194 svc_rc = SECSVC_DROP;
2199 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2200 svc_rc = SECSVC_DROP;
2207 * Used by ptlrpc server, to perform transformation upon request message of
2208 * incoming \a req. This must be the first thing to do with an incoming
2209 * request in ptlrpc layer.
2211 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2212 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2213 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2214 * reply message has been prepared.
2215 * \retval SECSVC_DROP failed, this request should be dropped.
2217 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2219 struct ptlrpc_sec_policy *policy;
2220 struct lustre_msg *msg = req->rq_reqbuf;
2226 LASSERT(req->rq_reqmsg == NULL);
2227 LASSERT(req->rq_repmsg == NULL);
2228 LASSERT(req->rq_svc_ctx == NULL);
2230 req->rq_req_swab_mask = 0;
2232 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2235 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2239 CERROR("error unpacking request from %s x%llu\n",
2240 libcfs_id2str(req->rq_peer), req->rq_xid);
2241 RETURN(SECSVC_DROP);
2244 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2245 req->rq_sp_from = LUSTRE_SP_ANY;
2246 req->rq_auth_uid = -1; /* set to INVALID_UID */
2247 req->rq_auth_mapped_uid = -1;
2249 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2251 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2252 RETURN(SECSVC_DROP);
2255 LASSERT(policy->sp_sops->accept);
2256 rc = policy->sp_sops->accept(req);
2257 sptlrpc_policy_put(policy);
2258 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2259 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2262 * if it's not null flavor (which means embedded packing msg),
2263 * reset the swab mask for the comming inner msg unpacking.
2265 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2266 req->rq_req_swab_mask = 0;
2268 /* sanity check for the request source */
2269 rc = sptlrpc_svc_check_from(req, rc);
2274 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2275 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2276 * a buffer of \a msglen size.
2278 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2280 struct ptlrpc_sec_policy *policy;
2281 struct ptlrpc_reply_state *rs;
2286 LASSERT(req->rq_svc_ctx);
2287 LASSERT(req->rq_svc_ctx->sc_policy);
2289 policy = req->rq_svc_ctx->sc_policy;
2290 LASSERT(policy->sp_sops->alloc_rs);
2292 rc = policy->sp_sops->alloc_rs(req, msglen);
2293 if (unlikely(rc == -ENOMEM)) {
2294 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2296 if (svcpt->scp_service->srv_max_reply_size <
2297 msglen + sizeof(struct ptlrpc_reply_state)) {
2298 /* Just return failure if the size is too big */
2299 CERROR("size of message is too big (%zd), %d allowed\n",
2300 msglen + sizeof(struct ptlrpc_reply_state),
2301 svcpt->scp_service->srv_max_reply_size);
2305 /* failed alloc, try emergency pool */
2306 rs = lustre_get_emerg_rs(svcpt);
2310 req->rq_reply_state = rs;
2311 rc = policy->sp_sops->alloc_rs(req, msglen);
2313 lustre_put_emerg_rs(rs);
2314 req->rq_reply_state = NULL;
2319 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2325 * Used by ptlrpc server, to perform transformation upon reply message.
2327 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2328 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2330 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2332 struct ptlrpc_sec_policy *policy;
2337 LASSERT(req->rq_svc_ctx);
2338 LASSERT(req->rq_svc_ctx->sc_policy);
2340 policy = req->rq_svc_ctx->sc_policy;
2341 LASSERT(policy->sp_sops->authorize);
2343 rc = policy->sp_sops->authorize(req);
2344 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2350 * Used by ptlrpc server, to free reply_state.
2352 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2354 struct ptlrpc_sec_policy *policy;
2355 unsigned int prealloc;
2359 LASSERT(rs->rs_svc_ctx);
2360 LASSERT(rs->rs_svc_ctx->sc_policy);
2362 policy = rs->rs_svc_ctx->sc_policy;
2363 LASSERT(policy->sp_sops->free_rs);
2365 prealloc = rs->rs_prealloc;
2366 policy->sp_sops->free_rs(rs);
2369 lustre_put_emerg_rs(rs);
2373 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2375 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2378 atomic_inc(&ctx->sc_refcount);
2381 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2383 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2388 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2389 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2390 if (ctx->sc_policy->sp_sops->free_ctx)
2391 ctx->sc_policy->sp_sops->free_ctx(ctx);
2393 req->rq_svc_ctx = NULL;
2396 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2398 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2403 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2404 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2405 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2407 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2414 * Perform transformation upon bulk data pointed by \a desc. This is called
2415 * before transforming the request message.
2417 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2418 struct ptlrpc_bulk_desc *desc)
2420 struct ptlrpc_cli_ctx *ctx;
2422 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2424 if (!req->rq_pack_bulk)
2427 ctx = req->rq_cli_ctx;
2428 if (ctx->cc_ops->wrap_bulk)
2429 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2432 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2435 * This is called after unwrap the reply message.
2436 * return nob of actual plain text size received, or error code.
2438 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2439 struct ptlrpc_bulk_desc *desc,
2442 struct ptlrpc_cli_ctx *ctx;
2445 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2447 if (!req->rq_pack_bulk)
2448 return desc->bd_nob_transferred;
2450 ctx = req->rq_cli_ctx;
2451 if (ctx->cc_ops->unwrap_bulk) {
2452 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2456 return desc->bd_nob_transferred;
2458 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2461 * This is called after unwrap the reply message.
2462 * return 0 for success or error code.
2464 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2465 struct ptlrpc_bulk_desc *desc)
2467 struct ptlrpc_cli_ctx *ctx;
2470 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2472 if (!req->rq_pack_bulk)
2475 ctx = req->rq_cli_ctx;
2476 if (ctx->cc_ops->unwrap_bulk) {
2477 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2483 * if everything is going right, nob should equals to nob_transferred.
2484 * in case of privacy mode, nob_transferred needs to be adjusted.
2486 if (desc->bd_nob != desc->bd_nob_transferred) {
2487 CERROR("nob %d doesn't match transferred nob %d\n",
2488 desc->bd_nob, desc->bd_nob_transferred);
2494 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2496 #ifdef HAVE_SERVER_SUPPORT
2498 * Performe transformation upon outgoing bulk read.
2500 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2501 struct ptlrpc_bulk_desc *desc)
2503 struct ptlrpc_svc_ctx *ctx;
2505 LASSERT(req->rq_bulk_read);
2507 if (!req->rq_pack_bulk)
2510 ctx = req->rq_svc_ctx;
2511 if (ctx->sc_policy->sp_sops->wrap_bulk)
2512 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2516 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2519 * Performe transformation upon incoming bulk write.
2521 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2522 struct ptlrpc_bulk_desc *desc)
2524 struct ptlrpc_svc_ctx *ctx;
2527 LASSERT(req->rq_bulk_write);
2530 * if it's in privacy mode, transferred should >= expected; otherwise
2531 * transferred should == expected.
2533 if (desc->bd_nob_transferred < desc->bd_nob ||
2534 (desc->bd_nob_transferred > desc->bd_nob &&
2535 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2536 SPTLRPC_BULK_SVC_PRIV)) {
2537 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2538 desc->bd_nob_transferred, desc->bd_nob);
2542 if (!req->rq_pack_bulk)
2545 ctx = req->rq_svc_ctx;
2546 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2547 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2549 CERROR("error unwrap bulk: %d\n", rc);
2552 /* return 0 to allow reply be sent */
2555 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2558 * Prepare buffers for incoming bulk write.
2560 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2561 struct ptlrpc_bulk_desc *desc)
2563 struct ptlrpc_svc_ctx *ctx;
2565 LASSERT(req->rq_bulk_write);
2567 if (!req->rq_pack_bulk)
2570 ctx = req->rq_svc_ctx;
2571 if (ctx->sc_policy->sp_sops->prep_bulk)
2572 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2576 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2578 #endif /* HAVE_SERVER_SUPPORT */
2581 * user descriptor helpers
2584 int sptlrpc_current_user_desc_size(void)
2588 ngroups = current_cred()->group_info->ngroups;
2590 if (ngroups > LUSTRE_MAX_GROUPS)
2591 ngroups = LUSTRE_MAX_GROUPS;
2592 return sptlrpc_user_desc_size(ngroups);
2594 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2596 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2598 struct ptlrpc_user_desc *pud;
2601 pud = lustre_msg_buf(msg, offset, 0);
2603 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2604 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2605 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2606 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2607 pud->pud_cap = cfs_curproc_cap_pack();
2608 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2611 ngroups = current_cred()->group_info->ngroups;
2612 if (pud->pud_ngroups > ngroups)
2613 pud->pud_ngroups = ngroups;
2614 #ifdef HAVE_GROUP_INFO_GID
2615 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2616 pud->pud_ngroups * sizeof(__u32));
2617 #else /* !HAVE_GROUP_INFO_GID */
2618 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2619 pud->pud_ngroups * sizeof(__u32));
2620 #endif /* HAVE_GROUP_INFO_GID */
2621 task_unlock(current);
2625 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2627 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2629 struct ptlrpc_user_desc *pud;
2632 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2637 __swab32s(&pud->pud_uid);
2638 __swab32s(&pud->pud_gid);
2639 __swab32s(&pud->pud_fsuid);
2640 __swab32s(&pud->pud_fsgid);
2641 __swab32s(&pud->pud_cap);
2642 __swab32s(&pud->pud_ngroups);
2645 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2646 CERROR("%u groups is too large\n", pud->pud_ngroups);
2650 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2651 msg->lm_buflens[offset]) {
2652 CERROR("%u groups are claimed but bufsize only %u\n",
2653 pud->pud_ngroups, msg->lm_buflens[offset]);
2658 for (i = 0; i < pud->pud_ngroups; i++)
2659 __swab32s(&pud->pud_groups[i]);
2664 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2670 const char *sec2target_str(struct ptlrpc_sec *sec)
2672 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2674 if (sec_is_reverse(sec))
2676 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2678 EXPORT_SYMBOL(sec2target_str);
2681 * return true if the bulk data is protected
2683 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2685 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2686 case SPTLRPC_BULK_SVC_INTG:
2687 case SPTLRPC_BULK_SVC_PRIV:
2693 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2696 * crypto API helper/alloc blkciper
2700 * initialize/finalize
2703 int sptlrpc_init(void)
2707 rwlock_init(&policy_lock);
2709 rc = sptlrpc_gc_init();
2713 rc = sptlrpc_conf_init();
2717 rc = sptlrpc_enc_pool_init();
2721 rc = sptlrpc_null_init();
2725 rc = sptlrpc_plain_init();
2729 rc = sptlrpc_lproc_init();
2736 sptlrpc_plain_fini();
2738 sptlrpc_null_fini();
2740 sptlrpc_enc_pool_fini();
2742 sptlrpc_conf_fini();
2749 void sptlrpc_fini(void)
2751 sptlrpc_lproc_fini();
2752 sptlrpc_plain_fini();
2753 sptlrpc_null_fini();
2754 sptlrpc_enc_pool_fini();
2755 sptlrpc_conf_fini();