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 /* Only output an error when the import is still active */
409 if (!test_bit(WORK_STRUCT_PENDING_BIT,
410 work_data_bits(&imp->imp_zombie_work)))
411 CERROR("import %p (%s) with no sec\n",
412 imp, ptlrpc_import_state_name(imp->imp_state));
416 if (unlikely((*sec)->ps_dying)) {
417 CERROR("attempt to use dying sec %p\n", sec);
418 sptlrpc_sec_put(*sec);
426 * Given a \a req, find or allocate an appropriate context for it.
427 * \pre req->rq_cli_ctx == NULL.
429 * \retval 0 succeed, and req->rq_cli_ctx is set.
430 * \retval -ev error number, and req->rq_cli_ctx == NULL.
432 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
434 struct obd_import *imp = req->rq_import;
435 struct ptlrpc_sec *sec;
440 LASSERT(!req->rq_cli_ctx);
443 rc = import_sec_validate_get(imp, &sec);
447 req->rq_cli_ctx = get_my_ctx(sec);
449 sptlrpc_sec_put(sec);
451 if (!req->rq_cli_ctx) {
452 CERROR("req %p: fail to get context\n", req);
453 RETURN(-ECONNREFUSED);
460 * Drop the context for \a req.
461 * \pre req->rq_cli_ctx != NULL.
462 * \post req->rq_cli_ctx == NULL.
464 * If \a sync == 0, this function should return quickly without sleep;
465 * otherwise it might trigger and wait for the whole process of sending
466 * an context-destroying rpc to server.
468 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
473 LASSERT(req->rq_cli_ctx);
476 * request might be asked to release earlier while still
477 * in the context waiting list.
479 if (!list_empty(&req->rq_ctx_chain)) {
480 spin_lock(&req->rq_cli_ctx->cc_lock);
481 list_del_init(&req->rq_ctx_chain);
482 spin_unlock(&req->rq_cli_ctx->cc_lock);
485 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
486 req->rq_cli_ctx = NULL;
491 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
492 struct ptlrpc_cli_ctx *oldctx,
493 struct ptlrpc_cli_ctx *newctx)
495 struct sptlrpc_flavor old_flvr;
496 char *reqmsg = NULL; /* to workaround old gcc */
500 LASSERT(req->rq_reqmsg);
501 LASSERT(req->rq_reqlen);
502 LASSERT(req->rq_replen);
505 "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
506 req, oldctx, oldctx->cc_vcred.vc_uid,
507 sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
508 sec2target_str(newctx->cc_sec), oldctx->cc_sec,
509 oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
510 newctx->cc_sec->ps_policy->sp_name);
513 old_flvr = req->rq_flvr;
515 /* save request message */
516 reqmsg_size = req->rq_reqlen;
517 if (reqmsg_size != 0) {
518 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
521 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
524 /* release old req/rep buf */
525 req->rq_cli_ctx = oldctx;
526 sptlrpc_cli_free_reqbuf(req);
527 sptlrpc_cli_free_repbuf(req);
528 req->rq_cli_ctx = newctx;
530 /* recalculate the flavor */
531 sptlrpc_req_set_flavor(req, 0);
534 * alloc new request buffer
535 * we don't need to alloc reply buffer here, leave it to the
536 * rest procedure of ptlrpc
538 if (reqmsg_size != 0) {
539 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
541 LASSERT(req->rq_reqmsg);
542 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
544 CWARN("failed to alloc reqbuf: %d\n", rc);
545 req->rq_flvr = old_flvr;
548 OBD_FREE_LARGE(reqmsg, reqmsg_size);
554 * If current context of \a req is dead somehow, e.g. we just switched flavor
555 * thus marked original contexts dead, we'll find a new context for it. if
556 * no switch is needed, \a req will end up with the same context.
558 * \note a request must have a context, to keep other parts of code happy.
559 * In any case of failure during the switching, we must restore the old one.
561 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
563 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
564 struct ptlrpc_cli_ctx *newctx;
571 sptlrpc_cli_ctx_get(oldctx);
572 sptlrpc_req_put_ctx(req, 0);
574 rc = sptlrpc_req_get_ctx(req);
576 LASSERT(!req->rq_cli_ctx);
578 /* restore old ctx */
579 req->rq_cli_ctx = oldctx;
583 newctx = req->rq_cli_ctx;
586 if (unlikely(newctx == oldctx &&
587 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
589 * still get the old dead ctx, usually means system too busy
592 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
593 newctx, newctx->cc_flags);
595 schedule_timeout_interruptible(cfs_time_seconds(1));
596 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
599 * new ctx not up to date yet
602 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
603 newctx, newctx->cc_flags);
606 * it's possible newctx == oldctx if we're switching
607 * subflavor with the same sec.
609 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
611 /* restore old ctx */
612 sptlrpc_req_put_ctx(req, 0);
613 req->rq_cli_ctx = oldctx;
617 LASSERT(req->rq_cli_ctx == newctx);
620 sptlrpc_cli_ctx_put(oldctx, 1);
623 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
626 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
628 if (cli_ctx_is_refreshed(ctx))
634 void ctx_refresh_interrupt(struct ptlrpc_request *req)
637 spin_lock(&req->rq_lock);
639 spin_unlock(&req->rq_lock);
643 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
645 spin_lock(&ctx->cc_lock);
646 if (!list_empty(&req->rq_ctx_chain))
647 list_del_init(&req->rq_ctx_chain);
648 spin_unlock(&ctx->cc_lock);
652 * To refresh the context of \req, if it's not up-to-date.
654 * - == 0: do not wait
655 * - == MAX_SCHEDULE_TIMEOUT: wait indefinitely
656 * - > 0: not supported
658 * The status of the context could be subject to be changed by other threads
659 * at any time. We allow this race, but once we return with 0, the caller will
660 * suppose it's uptodated and keep using it until the owning rpc is done.
662 * \retval 0 only if the context is uptodated.
663 * \retval -ev error number.
665 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
667 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
668 struct ptlrpc_sec *sec;
675 if (req->rq_ctx_init || req->rq_ctx_fini)
678 if (timeout != 0 && timeout != MAX_SCHEDULE_TIMEOUT) {
679 CERROR("req %p: invalid timeout %lu\n", req, timeout);
684 * during the process a request's context might change type even
685 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
689 rc = import_sec_validate_get(req->rq_import, &sec);
693 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
694 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
695 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
696 req_off_ctx_list(req, ctx);
697 sptlrpc_req_replace_dead_ctx(req);
698 ctx = req->rq_cli_ctx;
700 sptlrpc_sec_put(sec);
702 if (cli_ctx_is_eternal(ctx))
705 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
706 if (ctx->cc_ops->refresh)
707 ctx->cc_ops->refresh(ctx);
709 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
711 LASSERT(ctx->cc_ops->validate);
712 if (ctx->cc_ops->validate(ctx) == 0) {
713 req_off_ctx_list(req, ctx);
717 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
718 spin_lock(&req->rq_lock);
720 spin_unlock(&req->rq_lock);
721 req_off_ctx_list(req, ctx);
726 * There's a subtle issue for resending RPCs, suppose following
728 * 1. the request was sent to server.
729 * 2. recovery was kicked start, after finished the request was
731 * 3. resend the request.
732 * 4. old reply from server received, we accept and verify the reply.
733 * this has to be success, otherwise the error will be aware
735 * 5. new reply from server received, dropped by LNet.
737 * Note the xid of old & new request is the same. We can't simply
738 * change xid for the resent request because the server replies on
739 * it for reply reconstruction.
741 * Commonly the original context should be uptodate because we
742 * have an expiry nice time; server will keep its context because
743 * we at least hold a ref of old context which prevent context
744 * from destroying RPC being sent. So server still can accept the
745 * request and finish the RPC. But if that's not the case:
746 * 1. If server side context has been trimmed, a NO_CONTEXT will
747 * be returned, gss_cli_ctx_verify/unseal will switch to new
749 * 2. Current context never be refreshed, then we are fine: we
750 * never really send request with old context before.
752 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
753 unlikely(req->rq_reqmsg) &&
754 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
755 req_off_ctx_list(req, ctx);
759 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
760 req_off_ctx_list(req, ctx);
762 * don't switch ctx if import was deactivated
764 if (req->rq_import->imp_deactive) {
765 spin_lock(&req->rq_lock);
767 spin_unlock(&req->rq_lock);
771 rc = sptlrpc_req_replace_dead_ctx(req);
773 LASSERT(ctx == req->rq_cli_ctx);
774 CERROR("req %p: failed to replace dead ctx %p: %d\n",
776 spin_lock(&req->rq_lock);
778 spin_unlock(&req->rq_lock);
782 ctx = req->rq_cli_ctx;
787 * Now we're sure this context is during upcall, add myself into
790 spin_lock(&ctx->cc_lock);
791 if (list_empty(&req->rq_ctx_chain))
792 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
793 spin_unlock(&ctx->cc_lock);
796 RETURN(-EWOULDBLOCK);
798 /* Clear any flags that may be present from previous sends */
799 LASSERT(req->rq_receiving_reply == 0);
800 spin_lock(&req->rq_lock);
802 req->rq_timedout = 0;
805 spin_unlock(&req->rq_lock);
807 /* by now we know that timeout value is MAX_SCHEDULE_TIMEOUT,
808 * so wait indefinitely with non-fatal signals blocked
810 if (l_wait_event_abortable(req->rq_reply_waitq,
811 ctx_check_refresh(ctx)) == -ERESTARTSYS) {
813 ctx_refresh_interrupt(req);
817 * following cases could lead us here:
818 * - successfully refreshed;
820 * - timedout, and we don't want recover from the failure;
821 * - timedout, and waked up upon recovery finished;
822 * - someone else mark this ctx dead by force;
823 * - someone invalidate the req and call ptlrpc_client_wake_req(),
824 * e.g. ptlrpc_abort_inflight();
826 if (!cli_ctx_is_refreshed(ctx)) {
827 /* timed out or interruptted */
828 req_off_ctx_list(req, ctx);
837 /* Bring ptlrpc_sec context up-to-date */
838 int sptlrpc_export_update_ctx(struct obd_export *exp)
840 struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
841 struct ptlrpc_sec *sec = NULL;
842 struct ptlrpc_cli_ctx *ctx = NULL;
846 sec = sptlrpc_import_sec_ref(imp);
848 ctx = get_my_ctx(sec);
849 sptlrpc_sec_put(sec);
853 if (ctx->cc_ops->refresh)
854 rc = ctx->cc_ops->refresh(ctx);
855 sptlrpc_cli_ctx_put(ctx, 1);
861 * Initialize flavor settings for \a req, according to \a opcode.
863 * \note this could be called in two situations:
864 * - new request from ptlrpc_pre_req(), with proper @opcode
865 * - old request which changed ctx in the middle, with @opcode == 0
867 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
869 struct ptlrpc_sec *sec;
871 LASSERT(req->rq_import);
872 LASSERT(req->rq_cli_ctx);
873 LASSERT(req->rq_cli_ctx->cc_sec);
874 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
876 /* special security flags according to opcode */
880 case MGS_CONFIG_READ:
882 req->rq_bulk_read = 1;
886 req->rq_bulk_write = 1;
889 req->rq_ctx_init = 1;
892 req->rq_ctx_fini = 1;
895 /* init/fini rpc won't be resend, so can't be here */
896 LASSERT(req->rq_ctx_init == 0);
897 LASSERT(req->rq_ctx_fini == 0);
899 /* cleanup flags, which should be recalculated */
900 req->rq_pack_udesc = 0;
901 req->rq_pack_bulk = 0;
905 sec = req->rq_cli_ctx->cc_sec;
907 spin_lock(&sec->ps_lock);
908 req->rq_flvr = sec->ps_flvr;
909 spin_unlock(&sec->ps_lock);
912 * force SVC_NULL for context initiation rpc, SVC_INTG for context
915 if (unlikely(req->rq_ctx_init))
916 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
917 else if (unlikely(req->rq_ctx_fini))
918 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
920 /* user descriptor flag, null security can't do it anyway */
921 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
922 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
923 req->rq_pack_udesc = 1;
925 /* bulk security flag */
926 if ((req->rq_bulk_read || req->rq_bulk_write) &&
927 sptlrpc_flavor_has_bulk(&req->rq_flvr))
928 req->rq_pack_bulk = 1;
931 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
933 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
936 LASSERT(req->rq_clrbuf);
937 if (req->rq_pool || !req->rq_reqbuf)
940 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
941 req->rq_reqbuf = NULL;
942 req->rq_reqbuf_len = 0;
946 * Given an import \a imp, check whether current user has a valid context
947 * or not. We may create a new context and try to refresh it, and try
948 * repeatedly try in case of non-fatal errors. Return 0 means success.
950 int sptlrpc_import_check_ctx(struct obd_import *imp)
952 struct ptlrpc_sec *sec;
953 struct ptlrpc_cli_ctx *ctx;
954 struct ptlrpc_request *req = NULL;
961 sec = sptlrpc_import_sec_ref(imp);
962 ctx = get_my_ctx(sec);
963 sptlrpc_sec_put(sec);
968 if (cli_ctx_is_eternal(ctx) ||
969 ctx->cc_ops->validate(ctx) == 0) {
970 sptlrpc_cli_ctx_put(ctx, 1);
974 if (cli_ctx_is_error(ctx)) {
975 sptlrpc_cli_ctx_put(ctx, 1);
979 req = ptlrpc_request_cache_alloc(GFP_NOFS);
983 ptlrpc_cli_req_init(req);
984 atomic_set(&req->rq_refcount, 10000);
986 req->rq_import = imp;
987 req->rq_flvr = sec->ps_flvr;
988 req->rq_cli_ctx = ctx;
990 rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
991 LASSERT(list_empty(&req->rq_ctx_chain));
992 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
993 ptlrpc_request_cache_free(req);
999 * Used by ptlrpc client, to perform the pre-defined security transformation
1000 * upon the request message of \a req. After this function called,
1001 * req->rq_reqmsg is still accessible as clear text.
1003 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1005 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1011 LASSERT(ctx->cc_sec);
1012 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1015 * we wrap bulk request here because now we can be sure
1016 * the context is uptodate.
1019 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1024 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1025 case SPTLRPC_SVC_NULL:
1026 case SPTLRPC_SVC_AUTH:
1027 case SPTLRPC_SVC_INTG:
1028 LASSERT(ctx->cc_ops->sign);
1029 rc = ctx->cc_ops->sign(ctx, req);
1031 case SPTLRPC_SVC_PRIV:
1032 LASSERT(ctx->cc_ops->seal);
1033 rc = ctx->cc_ops->seal(ctx, req);
1040 LASSERT(req->rq_reqdata_len);
1041 LASSERT(req->rq_reqdata_len % 8 == 0);
1042 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1048 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1050 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1056 LASSERT(ctx->cc_sec);
1057 LASSERT(req->rq_repbuf);
1058 LASSERT(req->rq_repdata);
1059 LASSERT(req->rq_repmsg == NULL);
1061 req->rq_rep_swab_mask = 0;
1063 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1066 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1070 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1074 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1075 CERROR("replied data length %d too small\n",
1076 req->rq_repdata_len);
1080 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1081 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1082 CERROR("reply policy %u doesn't match request policy %u\n",
1083 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1084 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1088 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1089 case SPTLRPC_SVC_NULL:
1090 case SPTLRPC_SVC_AUTH:
1091 case SPTLRPC_SVC_INTG:
1092 LASSERT(ctx->cc_ops->verify);
1093 rc = ctx->cc_ops->verify(ctx, req);
1095 case SPTLRPC_SVC_PRIV:
1096 LASSERT(ctx->cc_ops->unseal);
1097 rc = ctx->cc_ops->unseal(ctx, req);
1102 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1104 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1106 req->rq_rep_swab_mask = 0;
1111 * Used by ptlrpc client, to perform security transformation upon the reply
1112 * message of \a req. After return successfully, req->rq_repmsg points to
1113 * the reply message in clear text.
1115 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1118 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1120 LASSERT(req->rq_repbuf);
1121 LASSERT(req->rq_repdata == NULL);
1122 LASSERT(req->rq_repmsg == NULL);
1123 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1125 if (req->rq_reply_off == 0 &&
1126 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1127 CERROR("real reply with offset 0\n");
1131 if (req->rq_reply_off % 8 != 0) {
1132 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1136 req->rq_repdata = (struct lustre_msg *)
1137 (req->rq_repbuf + req->rq_reply_off);
1138 req->rq_repdata_len = req->rq_nob_received;
1140 return do_cli_unwrap_reply(req);
1144 * Used by ptlrpc client, to perform security transformation upon the early
1145 * reply message of \a req. We expect the rq_reply_off is 0, and
1146 * rq_nob_received is the early reply size.
1148 * Because the receive buffer might be still posted, the reply data might be
1149 * changed at any time, no matter we're holding rq_lock or not. For this reason
1150 * we allocate a separate ptlrpc_request and reply buffer for early reply
1153 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1154 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1155 * \a *req_ret to release it.
1156 * \retval -ev error number, and \a req_ret will not be set.
1158 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1159 struct ptlrpc_request **req_ret)
1161 struct ptlrpc_request *early_req;
1163 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 "unwrap early reply: rc = %d", 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);
1277 * client side high-level security APIs
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;
1360 LASSERT(imp->imp_dlm_fake == 1);
1362 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1363 imp->imp_obd->obd_type->typ_name,
1364 imp->imp_obd->obd_name,
1365 sptlrpc_flavor2name(sf, str, sizeof(str)));
1367 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1368 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1370 LASSERT(imp->imp_dlm_fake == 0);
1372 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1373 imp->imp_obd->obd_type->typ_name,
1374 imp->imp_obd->obd_name,
1375 sptlrpc_flavor2name(sf, str, sizeof(str)));
1377 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1379 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1384 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1386 atomic_inc(&sec->ps_refcount);
1390 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1391 sptlrpc_gc_add_sec(sec);
1393 sptlrpc_policy_put(policy);
1399 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1401 struct ptlrpc_sec *sec;
1403 read_lock(&imp->imp_sec_lock);
1404 sec = sptlrpc_sec_get(imp->imp_sec);
1405 read_unlock(&imp->imp_sec_lock);
1409 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1411 static void sptlrpc_import_sec_install(struct obd_import *imp,
1412 struct ptlrpc_sec *sec)
1414 struct ptlrpc_sec *old_sec;
1416 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1418 write_lock(&imp->imp_sec_lock);
1419 old_sec = imp->imp_sec;
1421 write_unlock(&imp->imp_sec_lock);
1424 sptlrpc_sec_kill(old_sec);
1426 /* balance the ref taken by this import */
1427 sptlrpc_sec_put(old_sec);
1432 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1434 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1438 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1444 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1445 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1447 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1448 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1450 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1451 struct ptlrpc_svc_ctx *svc_ctx,
1452 struct sptlrpc_flavor *flvr)
1454 struct ptlrpc_connection *conn;
1455 struct sptlrpc_flavor sf;
1456 struct ptlrpc_sec *sec, *newsec;
1457 enum lustre_sec_part sp;
1468 conn = imp->imp_connection;
1470 if (svc_ctx == NULL) {
1471 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1473 * normal import, determine flavor from rule set, except
1474 * for mgc the flavor is predetermined.
1476 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1477 sf = cliobd->cl_flvr_mgc;
1479 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1481 &cliobd->cl_target_uuid,
1484 sp = imp->imp_obd->u.cli.cl_sp_me;
1486 /* reverse import, determine flavor from incoming reqeust */
1489 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1490 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1491 PTLRPC_SEC_FL_ROOTONLY;
1493 sp = sptlrpc_target_sec_part(imp->imp_obd);
1496 sec = sptlrpc_import_sec_ref(imp);
1500 if (flavor_equal(&sf, &sec->ps_flvr))
1503 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1504 imp->imp_obd->obd_name,
1505 obd_uuid2str(&conn->c_remote_uuid),
1506 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1507 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1508 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1509 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1510 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1511 imp->imp_obd->obd_name,
1512 obd_uuid2str(&conn->c_remote_uuid),
1513 LNET_NIDNET(conn->c_self),
1514 sptlrpc_flavor2name(&sf, str, sizeof(str)));
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));
1528 sptlrpc_sec_put(sec);
1532 void sptlrpc_import_sec_put(struct obd_import *imp)
1535 sptlrpc_sec_kill(imp->imp_sec);
1537 sptlrpc_sec_put(imp->imp_sec);
1538 imp->imp_sec = NULL;
1542 static void import_flush_ctx_common(struct obd_import *imp,
1543 uid_t uid, int grace, int force)
1545 struct ptlrpc_sec *sec;
1550 sec = sptlrpc_import_sec_ref(imp);
1554 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1555 sptlrpc_sec_put(sec);
1558 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()
1564 import_flush_ctx_common(imp, 0, 1, 1);
1567 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1569 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1572 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1574 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1576 import_flush_ctx_common(imp, -1, 1, 1);
1578 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1581 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1582 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1584 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1586 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1587 struct ptlrpc_sec_policy *policy;
1591 LASSERT(ctx->cc_sec);
1592 LASSERT(ctx->cc_sec->ps_policy);
1593 LASSERT(req->rq_reqmsg == NULL);
1594 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1596 policy = ctx->cc_sec->ps_policy;
1597 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1599 LASSERT(req->rq_reqmsg);
1600 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1602 /* zeroing preallocated buffer */
1604 memset(req->rq_reqmsg, 0, msgsize);
1611 * Used by ptlrpc client to free request buffer of \a req. After this
1612 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1614 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1616 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1617 struct ptlrpc_sec_policy *policy;
1620 LASSERT(ctx->cc_sec);
1621 LASSERT(ctx->cc_sec->ps_policy);
1622 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1624 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1627 policy = ctx->cc_sec->ps_policy;
1628 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1629 req->rq_reqmsg = NULL;
1633 * NOTE caller must guarantee the buffer size is enough for the enlargement
1635 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1636 int segment, int newsize)
1639 int oldsize, oldmsg_size, movesize;
1641 LASSERT(segment < msg->lm_bufcount);
1642 LASSERT(msg->lm_buflens[segment] <= newsize);
1644 if (msg->lm_buflens[segment] == newsize)
1647 /* nothing to do if we are enlarging the last segment */
1648 if (segment == msg->lm_bufcount - 1) {
1649 msg->lm_buflens[segment] = newsize;
1653 oldsize = msg->lm_buflens[segment];
1655 src = lustre_msg_buf(msg, segment + 1, 0);
1656 msg->lm_buflens[segment] = newsize;
1657 dst = lustre_msg_buf(msg, segment + 1, 0);
1658 msg->lm_buflens[segment] = oldsize;
1660 /* move from segment + 1 to end segment */
1661 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1662 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1663 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1664 LASSERT(movesize >= 0);
1667 memmove(dst, src, movesize);
1669 /* note we don't clear the ares where old data live, not secret */
1671 /* finally set new segment size */
1672 msg->lm_buflens[segment] = newsize;
1674 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1677 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1678 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1679 * preserved after the enlargement. this must be called after original request
1680 * buffer being allocated.
1682 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1683 * so caller should refresh its local pointers if needed.
1685 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1686 const struct req_msg_field *field,
1689 struct req_capsule *pill = &req->rq_pill;
1690 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1691 struct ptlrpc_sec_cops *cops;
1692 struct lustre_msg *msg = req->rq_reqmsg;
1693 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1697 LASSERT(msg->lm_bufcount > segment);
1698 LASSERT(msg->lm_buflens[segment] <= newsize);
1700 if (msg->lm_buflens[segment] == newsize)
1703 cops = ctx->cc_sec->ps_policy->sp_cops;
1704 LASSERT(cops->enlarge_reqbuf);
1705 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1707 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1710 * Used by ptlrpc client to allocate reply buffer of \a req.
1712 * \note After this, req->rq_repmsg is still not accessible.
1714 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1716 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1717 struct ptlrpc_sec_policy *policy;
1722 LASSERT(ctx->cc_sec);
1723 LASSERT(ctx->cc_sec->ps_policy);
1728 policy = ctx->cc_sec->ps_policy;
1729 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1733 * Used by ptlrpc client to free reply buffer of \a req. After this
1734 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1736 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1738 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1739 struct ptlrpc_sec_policy *policy;
1744 LASSERT(ctx->cc_sec);
1745 LASSERT(ctx->cc_sec->ps_policy);
1746 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1748 if (req->rq_repbuf == NULL)
1750 LASSERT(req->rq_repbuf_len);
1752 policy = ctx->cc_sec->ps_policy;
1753 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1754 req->rq_repmsg = NULL;
1757 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1759 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1760 struct ptlrpc_cli_ctx *ctx)
1762 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1764 if (!policy->sp_cops->install_rctx)
1766 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1769 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1770 struct ptlrpc_svc_ctx *ctx)
1772 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1774 if (!policy->sp_sops->install_rctx)
1776 return policy->sp_sops->install_rctx(imp, ctx);
1779 /* Get SELinux policy info from userspace */
1780 static int sepol_helper(struct obd_import *imp)
1782 char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1784 [0] = "/usr/sbin/l_getsepol",
1786 [2] = NULL, /* obd type */
1788 [4] = NULL, /* obd name */
1790 [6] = mtime_str, /* policy mtime */
1792 [8] = mode_str, /* enforcing mode */
1797 [1] = "PATH=/sbin:/usr/sbin",
1803 if (imp == NULL || imp->imp_obd == NULL ||
1804 imp->imp_obd->obd_type == NULL) {
1807 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1808 argv[4] = imp->imp_obd->obd_name;
1809 spin_lock(&imp->imp_sec->ps_lock);
1810 if (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1811 imp->imp_sec->ps_sepol[0] == '\0') {
1812 /* ps_sepol has not been initialized */
1818 mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1819 snprintf(mtime_str, sizeof(mtime_str), "%lld",
1820 mtime_ms / MSEC_PER_SEC);
1821 mode_str[0] = imp->imp_sec->ps_sepol[0];
1823 spin_unlock(&imp->imp_sec->ps_lock);
1824 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1831 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1835 if (send_sepol == 0 || !selinux_is_enabled())
1838 if (send_sepol == -1)
1839 /* send_sepol == -1 means fetch sepol status every time */
1842 spin_lock(&imp_sec->ps_lock);
1843 checknext = imp_sec->ps_sepol_checknext;
1844 spin_unlock(&imp_sec->ps_lock);
1846 /* next check is too far in time, please update */
1847 if (ktime_after(checknext,
1848 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1851 if (ktime_before(ktime_get(), checknext))
1852 /* too early to fetch sepol status */
1856 /* define new sepol_checknext time */
1857 spin_lock(&imp_sec->ps_lock);
1858 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1859 ktime_set(send_sepol, 0));
1860 spin_unlock(&imp_sec->ps_lock);
1865 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1867 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1872 (req->rq_sepol)[0] = '\0';
1874 #ifndef HAVE_SELINUX
1875 if (unlikely(send_sepol != 0))
1877 "Client cannot report SELinux status, it was not built against libselinux.\n");
1881 if (send_sepol == 0 || !selinux_is_enabled())
1884 if (imp_sec == NULL)
1887 /* Retrieve SELinux status info */
1888 if (sptlrpc_sepol_needs_check(imp_sec))
1889 rc = sepol_helper(req->rq_import);
1890 if (likely(rc == 0)) {
1891 spin_lock(&imp_sec->ps_lock);
1892 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1893 sizeof(req->rq_sepol));
1894 spin_unlock(&imp_sec->ps_lock);
1899 EXPORT_SYMBOL(sptlrpc_get_sepol);
1902 * server side security
1905 static int flavor_allowed(struct sptlrpc_flavor *exp,
1906 struct ptlrpc_request *req)
1908 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1910 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1913 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1914 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1915 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1916 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1922 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1925 * Given an export \a exp, check whether the flavor of incoming \a req
1926 * is allowed by the export \a exp. Main logic is about taking care of
1927 * changing configurations. Return 0 means success.
1929 int sptlrpc_target_export_check(struct obd_export *exp,
1930 struct ptlrpc_request *req)
1932 struct sptlrpc_flavor flavor;
1938 * client side export has no imp_reverse, skip
1939 * FIXME maybe we should check flavor this as well???
1941 if (exp->exp_imp_reverse == NULL)
1944 /* don't care about ctx fini rpc */
1945 if (req->rq_ctx_fini)
1948 spin_lock(&exp->exp_lock);
1951 * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1952 * the first req with the new flavor, then treat it as current flavor,
1953 * adapt reverse sec according to it.
1954 * note the first rpc with new flavor might not be with root ctx, in
1955 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1957 if (unlikely(exp->exp_flvr_changed) &&
1958 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1960 * make the new flavor as "current", and old ones as
1963 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1964 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1965 flavor = exp->exp_flvr_old[1];
1966 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1967 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1968 exp->exp_flvr_old[0] = exp->exp_flvr;
1969 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1970 EXP_FLVR_UPDATE_EXPIRE;
1971 exp->exp_flvr = flavor;
1973 /* flavor change finished */
1974 exp->exp_flvr_changed = 0;
1975 LASSERT(exp->exp_flvr_adapt == 1);
1977 /* if it's gss, we only interested in root ctx init */
1978 if (req->rq_auth_gss &&
1979 !(req->rq_ctx_init &&
1980 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1981 req->rq_auth_usr_ost))) {
1982 spin_unlock(&exp->exp_lock);
1983 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1984 req->rq_auth_gss, req->rq_ctx_init,
1985 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1986 req->rq_auth_usr_ost);
1990 exp->exp_flvr_adapt = 0;
1991 spin_unlock(&exp->exp_lock);
1993 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1994 req->rq_svc_ctx, &flavor);
1998 * if it equals to the current flavor, we accept it, but need to
1999 * dealing with reverse sec/ctx
2001 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
2003 * most cases should return here, we only interested in
2006 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2007 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2008 !req->rq_auth_usr_ost)) {
2009 spin_unlock(&exp->exp_lock);
2014 * if flavor just changed, we should not proceed, just leave
2015 * it and current flavor will be discovered and replaced
2016 * shortly, and let _this_ rpc pass through
2018 if (exp->exp_flvr_changed) {
2019 LASSERT(exp->exp_flvr_adapt);
2020 spin_unlock(&exp->exp_lock);
2024 if (exp->exp_flvr_adapt) {
2025 exp->exp_flvr_adapt = 0;
2026 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2027 exp, exp->exp_flvr.sf_rpc,
2028 exp->exp_flvr_old[0].sf_rpc,
2029 exp->exp_flvr_old[1].sf_rpc);
2030 flavor = exp->exp_flvr;
2031 spin_unlock(&exp->exp_lock);
2033 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2038 "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2039 exp, exp->exp_flvr.sf_rpc,
2040 exp->exp_flvr_old[0].sf_rpc,
2041 exp->exp_flvr_old[1].sf_rpc);
2042 spin_unlock(&exp->exp_lock);
2044 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2049 if (exp->exp_flvr_expire[0]) {
2050 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2051 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2053 "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2054 exp, exp->exp_flvr.sf_rpc,
2055 exp->exp_flvr_old[0].sf_rpc,
2056 exp->exp_flvr_old[1].sf_rpc,
2057 (s64)(exp->exp_flvr_expire[0] -
2058 ktime_get_real_seconds()));
2059 spin_unlock(&exp->exp_lock);
2063 CDEBUG(D_SEC, "mark middle expired\n");
2064 exp->exp_flvr_expire[0] = 0;
2066 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2067 exp->exp_flvr.sf_rpc,
2068 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2069 req->rq_flvr.sf_rpc);
2073 * now it doesn't match the current flavor, the only chance we can
2074 * accept it is match the old flavors which is not expired.
2076 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2077 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2078 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2079 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2081 exp->exp_flvr.sf_rpc,
2082 exp->exp_flvr_old[0].sf_rpc,
2083 exp->exp_flvr_old[1].sf_rpc,
2084 (s64)(exp->exp_flvr_expire[1] -
2085 ktime_get_real_seconds()));
2086 spin_unlock(&exp->exp_lock);
2090 CDEBUG(D_SEC, "mark oldest expired\n");
2091 exp->exp_flvr_expire[1] = 0;
2093 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2094 exp, exp->exp_flvr.sf_rpc,
2095 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2096 req->rq_flvr.sf_rpc);
2098 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2099 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2100 exp->exp_flvr_old[1].sf_rpc);
2103 spin_unlock(&exp->exp_lock);
2105 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2106 exp, exp->exp_obd->obd_name,
2107 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2108 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2109 req->rq_flvr.sf_rpc,
2110 exp->exp_flvr.sf_rpc,
2111 exp->exp_flvr_old[0].sf_rpc,
2112 exp->exp_flvr_expire[0] ?
2113 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2114 exp->exp_flvr_old[1].sf_rpc,
2115 exp->exp_flvr_expire[1] ?
2116 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2119 EXPORT_SYMBOL(sptlrpc_target_export_check);
2121 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2122 struct sptlrpc_rule_set *rset)
2124 struct obd_export *exp;
2125 struct sptlrpc_flavor new_flvr;
2129 spin_lock(&obd->obd_dev_lock);
2131 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2132 if (exp->exp_connection == NULL)
2136 * note if this export had just been updated flavor
2137 * (exp_flvr_changed == 1), this will override the
2140 spin_lock(&exp->exp_lock);
2141 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2142 exp->exp_connection->c_peer.nid,
2144 if (exp->exp_flvr_changed ||
2145 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2146 exp->exp_flvr_old[1] = new_flvr;
2147 exp->exp_flvr_expire[1] = 0;
2148 exp->exp_flvr_changed = 1;
2149 exp->exp_flvr_adapt = 1;
2151 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2152 exp, sptlrpc_part2name(exp->exp_sp_peer),
2153 exp->exp_flvr.sf_rpc,
2154 exp->exp_flvr_old[1].sf_rpc);
2156 spin_unlock(&exp->exp_lock);
2159 spin_unlock(&obd->obd_dev_lock);
2161 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2163 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2165 /* peer's claim is unreliable unless gss is being used */
2166 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2169 switch (req->rq_sp_from) {
2171 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2172 /* The below message is checked in sanity-sec test_33 */
2173 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2174 svc_rc = SECSVC_DROP;
2178 if (!req->rq_auth_usr_mdt) {
2179 /* The below message is checked in sanity-sec test_33 */
2180 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2181 svc_rc = SECSVC_DROP;
2185 if (!req->rq_auth_usr_ost) {
2186 /* The below message is checked in sanity-sec test_33 */
2187 DEBUG_REQ(D_ERROR, req, "faked source OST");
2188 svc_rc = SECSVC_DROP;
2193 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2194 !req->rq_auth_usr_ost) {
2195 /* The below message is checked in sanity-sec test_33 */
2196 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2197 svc_rc = SECSVC_DROP;
2202 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2203 svc_rc = SECSVC_DROP;
2210 * Used by ptlrpc server, to perform transformation upon request message of
2211 * incoming \a req. This must be the first thing to do with an incoming
2212 * request in ptlrpc layer.
2214 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2215 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2216 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2217 * reply message has been prepared.
2218 * \retval SECSVC_DROP failed, this request should be dropped.
2220 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2222 struct ptlrpc_sec_policy *policy;
2223 struct lustre_msg *msg = req->rq_reqbuf;
2229 LASSERT(req->rq_reqmsg == NULL);
2230 LASSERT(req->rq_repmsg == NULL);
2231 LASSERT(req->rq_svc_ctx == NULL);
2233 req->rq_req_swab_mask = 0;
2235 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2238 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2242 CERROR("error unpacking request from %s x%llu\n",
2243 libcfs_id2str(req->rq_peer), req->rq_xid);
2244 RETURN(SECSVC_DROP);
2247 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2248 req->rq_sp_from = LUSTRE_SP_ANY;
2249 req->rq_auth_uid = -1; /* set to INVALID_UID */
2250 req->rq_auth_mapped_uid = -1;
2252 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2254 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2255 RETURN(SECSVC_DROP);
2258 LASSERT(policy->sp_sops->accept);
2259 rc = policy->sp_sops->accept(req);
2260 sptlrpc_policy_put(policy);
2261 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2262 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2265 * if it's not null flavor (which means embedded packing msg),
2266 * reset the swab mask for the comming inner msg unpacking.
2268 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2269 req->rq_req_swab_mask = 0;
2271 /* sanity check for the request source */
2272 rc = sptlrpc_svc_check_from(req, rc);
2277 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2278 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2279 * a buffer of \a msglen size.
2281 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2283 struct ptlrpc_sec_policy *policy;
2284 struct ptlrpc_reply_state *rs;
2289 LASSERT(req->rq_svc_ctx);
2290 LASSERT(req->rq_svc_ctx->sc_policy);
2292 policy = req->rq_svc_ctx->sc_policy;
2293 LASSERT(policy->sp_sops->alloc_rs);
2295 rc = policy->sp_sops->alloc_rs(req, msglen);
2296 if (unlikely(rc == -ENOMEM)) {
2297 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2299 if (svcpt->scp_service->srv_max_reply_size <
2300 msglen + sizeof(struct ptlrpc_reply_state)) {
2301 /* Just return failure if the size is too big */
2302 CERROR("size of message is too big (%zd), %d allowed\n",
2303 msglen + sizeof(struct ptlrpc_reply_state),
2304 svcpt->scp_service->srv_max_reply_size);
2308 /* failed alloc, try emergency pool */
2309 rs = lustre_get_emerg_rs(svcpt);
2313 req->rq_reply_state = rs;
2314 rc = policy->sp_sops->alloc_rs(req, msglen);
2316 lustre_put_emerg_rs(rs);
2317 req->rq_reply_state = NULL;
2322 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2328 * Used by ptlrpc server, to perform transformation upon reply message.
2330 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2331 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2333 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2335 struct ptlrpc_sec_policy *policy;
2340 LASSERT(req->rq_svc_ctx);
2341 LASSERT(req->rq_svc_ctx->sc_policy);
2343 policy = req->rq_svc_ctx->sc_policy;
2344 LASSERT(policy->sp_sops->authorize);
2346 rc = policy->sp_sops->authorize(req);
2347 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2353 * Used by ptlrpc server, to free reply_state.
2355 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2357 struct ptlrpc_sec_policy *policy;
2358 unsigned int prealloc;
2362 LASSERT(rs->rs_svc_ctx);
2363 LASSERT(rs->rs_svc_ctx->sc_policy);
2365 policy = rs->rs_svc_ctx->sc_policy;
2366 LASSERT(policy->sp_sops->free_rs);
2368 prealloc = rs->rs_prealloc;
2369 policy->sp_sops->free_rs(rs);
2372 lustre_put_emerg_rs(rs);
2376 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2378 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2381 atomic_inc(&ctx->sc_refcount);
2384 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2386 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2391 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2392 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2393 if (ctx->sc_policy->sp_sops->free_ctx)
2394 ctx->sc_policy->sp_sops->free_ctx(ctx);
2396 req->rq_svc_ctx = NULL;
2399 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2401 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2406 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2407 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2408 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2410 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2417 * Perform transformation upon bulk data pointed by \a desc. This is called
2418 * before transforming the request message.
2420 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2421 struct ptlrpc_bulk_desc *desc)
2423 struct ptlrpc_cli_ctx *ctx;
2425 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2427 if (!req->rq_pack_bulk)
2430 ctx = req->rq_cli_ctx;
2431 if (ctx->cc_ops->wrap_bulk)
2432 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2435 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2438 * This is called after unwrap the reply message.
2439 * return nob of actual plain text size received, or error code.
2441 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2442 struct ptlrpc_bulk_desc *desc,
2445 struct ptlrpc_cli_ctx *ctx;
2448 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2450 if (!req->rq_pack_bulk)
2451 return desc->bd_nob_transferred;
2453 ctx = req->rq_cli_ctx;
2454 if (ctx->cc_ops->unwrap_bulk) {
2455 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2459 return desc->bd_nob_transferred;
2461 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2464 * This is called after unwrap the reply message.
2465 * return 0 for success or error code.
2467 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2468 struct ptlrpc_bulk_desc *desc)
2470 struct ptlrpc_cli_ctx *ctx;
2473 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2475 if (!req->rq_pack_bulk)
2478 ctx = req->rq_cli_ctx;
2479 if (ctx->cc_ops->unwrap_bulk) {
2480 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2486 * if everything is going right, nob should equals to nob_transferred.
2487 * in case of privacy mode, nob_transferred needs to be adjusted.
2489 if (desc->bd_nob != desc->bd_nob_transferred) {
2490 CERROR("nob %d doesn't match transferred nob %d\n",
2491 desc->bd_nob, desc->bd_nob_transferred);
2497 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2499 #ifdef HAVE_SERVER_SUPPORT
2501 * Performe transformation upon outgoing bulk read.
2503 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2504 struct ptlrpc_bulk_desc *desc)
2506 struct ptlrpc_svc_ctx *ctx;
2508 LASSERT(req->rq_bulk_read);
2510 if (!req->rq_pack_bulk)
2513 ctx = req->rq_svc_ctx;
2514 if (ctx->sc_policy->sp_sops->wrap_bulk)
2515 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2519 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2522 * Performe transformation upon incoming bulk write.
2524 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2525 struct ptlrpc_bulk_desc *desc)
2527 struct ptlrpc_svc_ctx *ctx;
2530 LASSERT(req->rq_bulk_write);
2533 * if it's in privacy mode, transferred should >= expected; otherwise
2534 * transferred should == expected.
2536 if (desc->bd_nob_transferred < desc->bd_nob ||
2537 (desc->bd_nob_transferred > desc->bd_nob &&
2538 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2539 SPTLRPC_BULK_SVC_PRIV)) {
2540 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2541 desc->bd_nob_transferred, desc->bd_nob);
2545 if (!req->rq_pack_bulk)
2548 ctx = req->rq_svc_ctx;
2549 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2550 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2552 CERROR("error unwrap bulk: %d\n", rc);
2555 /* return 0 to allow reply be sent */
2558 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2561 * Prepare buffers for incoming bulk write.
2563 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2564 struct ptlrpc_bulk_desc *desc)
2566 struct ptlrpc_svc_ctx *ctx;
2568 LASSERT(req->rq_bulk_write);
2570 if (!req->rq_pack_bulk)
2573 ctx = req->rq_svc_ctx;
2574 if (ctx->sc_policy->sp_sops->prep_bulk)
2575 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2579 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2581 #endif /* HAVE_SERVER_SUPPORT */
2584 * user descriptor helpers
2587 int sptlrpc_current_user_desc_size(void)
2591 ngroups = current_cred()->group_info->ngroups;
2593 if (ngroups > LUSTRE_MAX_GROUPS)
2594 ngroups = LUSTRE_MAX_GROUPS;
2595 return sptlrpc_user_desc_size(ngroups);
2597 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2599 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2601 struct ptlrpc_user_desc *pud;
2604 pud = lustre_msg_buf(msg, offset, 0);
2606 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2607 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2608 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2609 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2610 pud->pud_cap = cfs_curproc_cap_pack();
2611 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2614 ngroups = current_cred()->group_info->ngroups;
2615 if (pud->pud_ngroups > ngroups)
2616 pud->pud_ngroups = ngroups;
2617 #ifdef HAVE_GROUP_INFO_GID
2618 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2619 pud->pud_ngroups * sizeof(__u32));
2620 #else /* !HAVE_GROUP_INFO_GID */
2621 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2622 pud->pud_ngroups * sizeof(__u32));
2623 #endif /* HAVE_GROUP_INFO_GID */
2624 task_unlock(current);
2628 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2630 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2632 struct ptlrpc_user_desc *pud;
2635 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2640 __swab32s(&pud->pud_uid);
2641 __swab32s(&pud->pud_gid);
2642 __swab32s(&pud->pud_fsuid);
2643 __swab32s(&pud->pud_fsgid);
2644 __swab32s(&pud->pud_cap);
2645 __swab32s(&pud->pud_ngroups);
2648 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2649 CERROR("%u groups is too large\n", pud->pud_ngroups);
2653 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2654 msg->lm_buflens[offset]) {
2655 CERROR("%u groups are claimed but bufsize only %u\n",
2656 pud->pud_ngroups, msg->lm_buflens[offset]);
2661 for (i = 0; i < pud->pud_ngroups; i++)
2662 __swab32s(&pud->pud_groups[i]);
2667 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2673 const char *sec2target_str(struct ptlrpc_sec *sec)
2675 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2677 if (sec_is_reverse(sec))
2679 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2681 EXPORT_SYMBOL(sec2target_str);
2684 * return true if the bulk data is protected
2686 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2688 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2689 case SPTLRPC_BULK_SVC_INTG:
2690 case SPTLRPC_BULK_SVC_PRIV:
2696 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2699 * crypto API helper/alloc blkciper
2703 * initialize/finalize
2706 int sptlrpc_init(void)
2710 rwlock_init(&policy_lock);
2712 rc = sptlrpc_gc_init();
2716 rc = sptlrpc_conf_init();
2720 rc = sptlrpc_enc_pool_init();
2724 rc = sptlrpc_null_init();
2728 rc = sptlrpc_plain_init();
2732 rc = sptlrpc_lproc_init();
2739 sptlrpc_plain_fini();
2741 sptlrpc_null_fini();
2743 sptlrpc_enc_pool_fini();
2745 sptlrpc_conf_fini();
2752 void sptlrpc_fini(void)
2754 sptlrpc_lproc_fini();
2755 sptlrpc_plain_fini();
2756 sptlrpc_null_fini();
2757 sptlrpc_enc_pool_fini();
2758 sptlrpc_conf_fini();