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 LASSERT(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);
835 * Initialize flavor settings for \a req, according to \a opcode.
837 * \note this could be called in two situations:
838 * - new request from ptlrpc_pre_req(), with proper @opcode
839 * - old request which changed ctx in the middle, with @opcode == 0
841 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
843 struct ptlrpc_sec *sec;
845 LASSERT(req->rq_import);
846 LASSERT(req->rq_cli_ctx);
847 LASSERT(req->rq_cli_ctx->cc_sec);
848 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
850 /* special security flags according to opcode */
854 case MGS_CONFIG_READ:
856 req->rq_bulk_read = 1;
860 req->rq_bulk_write = 1;
863 req->rq_ctx_init = 1;
866 req->rq_ctx_fini = 1;
869 /* init/fini rpc won't be resend, so can't be here */
870 LASSERT(req->rq_ctx_init == 0);
871 LASSERT(req->rq_ctx_fini == 0);
873 /* cleanup flags, which should be recalculated */
874 req->rq_pack_udesc = 0;
875 req->rq_pack_bulk = 0;
879 sec = req->rq_cli_ctx->cc_sec;
881 spin_lock(&sec->ps_lock);
882 req->rq_flvr = sec->ps_flvr;
883 spin_unlock(&sec->ps_lock);
886 * force SVC_NULL for context initiation rpc, SVC_INTG for context
889 if (unlikely(req->rq_ctx_init))
890 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
891 else if (unlikely(req->rq_ctx_fini))
892 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
894 /* user descriptor flag, null security can't do it anyway */
895 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
896 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
897 req->rq_pack_udesc = 1;
899 /* bulk security flag */
900 if ((req->rq_bulk_read || req->rq_bulk_write) &&
901 sptlrpc_flavor_has_bulk(&req->rq_flvr))
902 req->rq_pack_bulk = 1;
905 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
907 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
910 LASSERT(req->rq_clrbuf);
911 if (req->rq_pool || !req->rq_reqbuf)
914 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
915 req->rq_reqbuf = NULL;
916 req->rq_reqbuf_len = 0;
920 * Given an import \a imp, check whether current user has a valid context
921 * or not. We may create a new context and try to refresh it, and try
922 * repeatedly try in case of non-fatal errors. Return 0 means success.
924 int sptlrpc_import_check_ctx(struct obd_import *imp)
926 struct ptlrpc_sec *sec;
927 struct ptlrpc_cli_ctx *ctx;
928 struct ptlrpc_request *req = NULL;
935 sec = sptlrpc_import_sec_ref(imp);
936 ctx = get_my_ctx(sec);
937 sptlrpc_sec_put(sec);
942 if (cli_ctx_is_eternal(ctx) ||
943 ctx->cc_ops->validate(ctx) == 0) {
944 sptlrpc_cli_ctx_put(ctx, 1);
948 if (cli_ctx_is_error(ctx)) {
949 sptlrpc_cli_ctx_put(ctx, 1);
953 req = ptlrpc_request_cache_alloc(GFP_NOFS);
957 ptlrpc_cli_req_init(req);
958 atomic_set(&req->rq_refcount, 10000);
960 req->rq_import = imp;
961 req->rq_flvr = sec->ps_flvr;
962 req->rq_cli_ctx = ctx;
964 rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
965 LASSERT(list_empty(&req->rq_ctx_chain));
966 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
967 ptlrpc_request_cache_free(req);
973 * Used by ptlrpc client, to perform the pre-defined security transformation
974 * upon the request message of \a req. After this function called,
975 * req->rq_reqmsg is still accessible as clear text.
977 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
979 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
985 LASSERT(ctx->cc_sec);
986 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
989 * we wrap bulk request here because now we can be sure
990 * the context is uptodate.
993 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
998 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
999 case SPTLRPC_SVC_NULL:
1000 case SPTLRPC_SVC_AUTH:
1001 case SPTLRPC_SVC_INTG:
1002 LASSERT(ctx->cc_ops->sign);
1003 rc = ctx->cc_ops->sign(ctx, req);
1005 case SPTLRPC_SVC_PRIV:
1006 LASSERT(ctx->cc_ops->seal);
1007 rc = ctx->cc_ops->seal(ctx, req);
1014 LASSERT(req->rq_reqdata_len);
1015 LASSERT(req->rq_reqdata_len % 8 == 0);
1016 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1022 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1024 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1030 LASSERT(ctx->cc_sec);
1031 LASSERT(req->rq_repbuf);
1032 LASSERT(req->rq_repdata);
1033 LASSERT(req->rq_repmsg == NULL);
1035 req->rq_rep_swab_mask = 0;
1037 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1040 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1044 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1048 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1049 CERROR("replied data length %d too small\n",
1050 req->rq_repdata_len);
1054 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1055 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1056 CERROR("reply policy %u doesn't match request policy %u\n",
1057 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1058 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1062 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1063 case SPTLRPC_SVC_NULL:
1064 case SPTLRPC_SVC_AUTH:
1065 case SPTLRPC_SVC_INTG:
1066 LASSERT(ctx->cc_ops->verify);
1067 rc = ctx->cc_ops->verify(ctx, req);
1069 case SPTLRPC_SVC_PRIV:
1070 LASSERT(ctx->cc_ops->unseal);
1071 rc = ctx->cc_ops->unseal(ctx, req);
1076 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1078 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1080 req->rq_rep_swab_mask = 0;
1085 * Used by ptlrpc client, to perform security transformation upon the reply
1086 * message of \a req. After return successfully, req->rq_repmsg points to
1087 * the reply message in clear text.
1089 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1092 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1094 LASSERT(req->rq_repbuf);
1095 LASSERT(req->rq_repdata == NULL);
1096 LASSERT(req->rq_repmsg == NULL);
1097 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1099 if (req->rq_reply_off == 0 &&
1100 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1101 CERROR("real reply with offset 0\n");
1105 if (req->rq_reply_off % 8 != 0) {
1106 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1110 req->rq_repdata = (struct lustre_msg *)
1111 (req->rq_repbuf + req->rq_reply_off);
1112 req->rq_repdata_len = req->rq_nob_received;
1114 return do_cli_unwrap_reply(req);
1118 * Used by ptlrpc client, to perform security transformation upon the early
1119 * reply message of \a req. We expect the rq_reply_off is 0, and
1120 * rq_nob_received is the early reply size.
1122 * Because the receive buffer might be still posted, the reply data might be
1123 * changed at any time, no matter we're holding rq_lock or not. For this reason
1124 * we allocate a separate ptlrpc_request and reply buffer for early reply
1127 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1128 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1129 * \a *req_ret to release it.
1130 * \retval -ev error number, and \a req_ret will not be set.
1132 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1133 struct ptlrpc_request **req_ret)
1135 struct ptlrpc_request *early_req;
1137 int early_bufsz, early_size;
1142 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1143 if (early_req == NULL)
1146 ptlrpc_cli_req_init(early_req);
1148 early_size = req->rq_nob_received;
1149 early_bufsz = size_roundup_power2(early_size);
1150 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1151 if (early_buf == NULL)
1152 GOTO(err_req, rc = -ENOMEM);
1154 /* sanity checkings and copy data out, do it inside spinlock */
1155 spin_lock(&req->rq_lock);
1157 if (req->rq_replied) {
1158 spin_unlock(&req->rq_lock);
1159 GOTO(err_buf, rc = -EALREADY);
1162 LASSERT(req->rq_repbuf);
1163 LASSERT(req->rq_repdata == NULL);
1164 LASSERT(req->rq_repmsg == NULL);
1166 if (req->rq_reply_off != 0) {
1167 CERROR("early reply with offset %u\n", req->rq_reply_off);
1168 spin_unlock(&req->rq_lock);
1169 GOTO(err_buf, rc = -EPROTO);
1172 if (req->rq_nob_received != early_size) {
1173 /* even another early arrived the size should be the same */
1174 CERROR("data size has changed from %u to %u\n",
1175 early_size, req->rq_nob_received);
1176 spin_unlock(&req->rq_lock);
1177 GOTO(err_buf, rc = -EINVAL);
1180 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1181 CERROR("early reply length %d too small\n",
1182 req->rq_nob_received);
1183 spin_unlock(&req->rq_lock);
1184 GOTO(err_buf, rc = -EALREADY);
1187 memcpy(early_buf, req->rq_repbuf, early_size);
1188 spin_unlock(&req->rq_lock);
1190 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1191 early_req->rq_flvr = req->rq_flvr;
1192 early_req->rq_repbuf = early_buf;
1193 early_req->rq_repbuf_len = early_bufsz;
1194 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1195 early_req->rq_repdata_len = early_size;
1196 early_req->rq_early = 1;
1197 early_req->rq_reqmsg = req->rq_reqmsg;
1199 rc = do_cli_unwrap_reply(early_req);
1201 DEBUG_REQ(D_ADAPTTO, early_req,
1202 "unwrap early reply: rc = %d", rc);
1206 LASSERT(early_req->rq_repmsg);
1207 *req_ret = early_req;
1211 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1213 OBD_FREE_LARGE(early_buf, early_bufsz);
1215 ptlrpc_request_cache_free(early_req);
1220 * Used by ptlrpc client, to release a processed early reply \a early_req.
1222 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1224 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1226 LASSERT(early_req->rq_repbuf);
1227 LASSERT(early_req->rq_repdata);
1228 LASSERT(early_req->rq_repmsg);
1230 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1231 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1232 ptlrpc_request_cache_free(early_req);
1235 /**************************************************
1237 **************************************************/
1240 * "fixed" sec (e.g. null) use sec_id < 0
1242 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1244 int sptlrpc_get_next_secid(void)
1246 return atomic_inc_return(&sptlrpc_sec_id);
1248 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1251 * client side high-level security APIs
1254 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1255 int grace, int force)
1257 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1259 LASSERT(policy->sp_cops);
1260 LASSERT(policy->sp_cops->flush_ctx_cache);
1262 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1265 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1267 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1269 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1270 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1271 LASSERT(policy->sp_cops->destroy_sec);
1273 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1275 policy->sp_cops->destroy_sec(sec);
1276 sptlrpc_policy_put(policy);
1279 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1281 sec_cop_destroy_sec(sec);
1283 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1285 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1287 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1289 if (sec->ps_policy->sp_cops->kill_sec) {
1290 sec->ps_policy->sp_cops->kill_sec(sec);
1292 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1296 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1299 atomic_inc(&sec->ps_refcount);
1303 EXPORT_SYMBOL(sptlrpc_sec_get);
1305 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1308 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1310 if (atomic_dec_and_test(&sec->ps_refcount)) {
1311 sptlrpc_gc_del_sec(sec);
1312 sec_cop_destroy_sec(sec);
1316 EXPORT_SYMBOL(sptlrpc_sec_put);
1319 * policy module is responsible for taking refrence of import
1322 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1323 struct ptlrpc_svc_ctx *svc_ctx,
1324 struct sptlrpc_flavor *sf,
1325 enum lustre_sec_part sp)
1327 struct ptlrpc_sec_policy *policy;
1328 struct ptlrpc_sec *sec;
1334 LASSERT(imp->imp_dlm_fake == 1);
1336 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1337 imp->imp_obd->obd_type->typ_name,
1338 imp->imp_obd->obd_name,
1339 sptlrpc_flavor2name(sf, str, sizeof(str)));
1341 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1342 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1344 LASSERT(imp->imp_dlm_fake == 0);
1346 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1347 imp->imp_obd->obd_type->typ_name,
1348 imp->imp_obd->obd_name,
1349 sptlrpc_flavor2name(sf, str, sizeof(str)));
1351 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1353 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1358 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1360 atomic_inc(&sec->ps_refcount);
1364 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1365 sptlrpc_gc_add_sec(sec);
1367 sptlrpc_policy_put(policy);
1373 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1375 struct ptlrpc_sec *sec;
1377 read_lock(&imp->imp_sec_lock);
1378 sec = sptlrpc_sec_get(imp->imp_sec);
1379 read_unlock(&imp->imp_sec_lock);
1383 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1385 static void sptlrpc_import_sec_install(struct obd_import *imp,
1386 struct ptlrpc_sec *sec)
1388 struct ptlrpc_sec *old_sec;
1390 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1392 write_lock(&imp->imp_sec_lock);
1393 old_sec = imp->imp_sec;
1395 write_unlock(&imp->imp_sec_lock);
1398 sptlrpc_sec_kill(old_sec);
1400 /* balance the ref taken by this import */
1401 sptlrpc_sec_put(old_sec);
1406 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1408 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1412 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1418 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1419 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1421 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1422 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1424 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1425 struct ptlrpc_svc_ctx *svc_ctx,
1426 struct sptlrpc_flavor *flvr)
1428 struct ptlrpc_connection *conn;
1429 struct sptlrpc_flavor sf;
1430 struct ptlrpc_sec *sec, *newsec;
1431 enum lustre_sec_part sp;
1442 conn = imp->imp_connection;
1444 if (svc_ctx == NULL) {
1445 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1447 * normal import, determine flavor from rule set, except
1448 * for mgc the flavor is predetermined.
1450 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1451 sf = cliobd->cl_flvr_mgc;
1453 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1455 &cliobd->cl_target_uuid,
1458 sp = imp->imp_obd->u.cli.cl_sp_me;
1460 /* reverse import, determine flavor from incoming reqeust */
1463 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1464 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1465 PTLRPC_SEC_FL_ROOTONLY;
1467 sp = sptlrpc_target_sec_part(imp->imp_obd);
1470 sec = sptlrpc_import_sec_ref(imp);
1474 if (flavor_equal(&sf, &sec->ps_flvr))
1477 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1478 imp->imp_obd->obd_name,
1479 obd_uuid2str(&conn->c_remote_uuid),
1480 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1481 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1482 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1483 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1484 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1485 imp->imp_obd->obd_name,
1486 obd_uuid2str(&conn->c_remote_uuid),
1487 LNET_NIDNET(conn->c_self),
1488 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1491 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1493 sptlrpc_import_sec_install(imp, newsec);
1495 CERROR("import %s->%s: failed to create new sec\n",
1496 imp->imp_obd->obd_name,
1497 obd_uuid2str(&conn->c_remote_uuid));
1502 sptlrpc_sec_put(sec);
1506 void sptlrpc_import_sec_put(struct obd_import *imp)
1509 sptlrpc_sec_kill(imp->imp_sec);
1511 sptlrpc_sec_put(imp->imp_sec);
1512 imp->imp_sec = NULL;
1516 static void import_flush_ctx_common(struct obd_import *imp,
1517 uid_t uid, int grace, int force)
1519 struct ptlrpc_sec *sec;
1524 sec = sptlrpc_import_sec_ref(imp);
1528 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1529 sptlrpc_sec_put(sec);
1532 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1535 * it's important to use grace mode, see explain in
1536 * sptlrpc_req_refresh_ctx()
1538 import_flush_ctx_common(imp, 0, 1, 1);
1541 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1543 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1546 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1548 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1550 import_flush_ctx_common(imp, -1, 1, 1);
1552 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1555 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1556 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1558 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1560 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1561 struct ptlrpc_sec_policy *policy;
1565 LASSERT(ctx->cc_sec);
1566 LASSERT(ctx->cc_sec->ps_policy);
1567 LASSERT(req->rq_reqmsg == NULL);
1568 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1570 policy = ctx->cc_sec->ps_policy;
1571 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1573 LASSERT(req->rq_reqmsg);
1574 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1576 /* zeroing preallocated buffer */
1578 memset(req->rq_reqmsg, 0, msgsize);
1585 * Used by ptlrpc client to free request buffer of \a req. After this
1586 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1588 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1590 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1591 struct ptlrpc_sec_policy *policy;
1594 LASSERT(ctx->cc_sec);
1595 LASSERT(ctx->cc_sec->ps_policy);
1596 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1598 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1601 policy = ctx->cc_sec->ps_policy;
1602 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1603 req->rq_reqmsg = NULL;
1607 * NOTE caller must guarantee the buffer size is enough for the enlargement
1609 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1610 int segment, int newsize)
1613 int oldsize, oldmsg_size, movesize;
1615 LASSERT(segment < msg->lm_bufcount);
1616 LASSERT(msg->lm_buflens[segment] <= newsize);
1618 if (msg->lm_buflens[segment] == newsize)
1621 /* nothing to do if we are enlarging the last segment */
1622 if (segment == msg->lm_bufcount - 1) {
1623 msg->lm_buflens[segment] = newsize;
1627 oldsize = msg->lm_buflens[segment];
1629 src = lustre_msg_buf(msg, segment + 1, 0);
1630 msg->lm_buflens[segment] = newsize;
1631 dst = lustre_msg_buf(msg, segment + 1, 0);
1632 msg->lm_buflens[segment] = oldsize;
1634 /* move from segment + 1 to end segment */
1635 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1636 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1637 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1638 LASSERT(movesize >= 0);
1641 memmove(dst, src, movesize);
1643 /* note we don't clear the ares where old data live, not secret */
1645 /* finally set new segment size */
1646 msg->lm_buflens[segment] = newsize;
1648 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1651 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1652 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1653 * preserved after the enlargement. this must be called after original request
1654 * buffer being allocated.
1656 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1657 * so caller should refresh its local pointers if needed.
1659 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1660 const struct req_msg_field *field,
1663 struct req_capsule *pill = &req->rq_pill;
1664 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1665 struct ptlrpc_sec_cops *cops;
1666 struct lustre_msg *msg = req->rq_reqmsg;
1667 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1671 LASSERT(msg->lm_bufcount > segment);
1672 LASSERT(msg->lm_buflens[segment] <= newsize);
1674 if (msg->lm_buflens[segment] == newsize)
1677 cops = ctx->cc_sec->ps_policy->sp_cops;
1678 LASSERT(cops->enlarge_reqbuf);
1679 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1681 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1684 * Used by ptlrpc client to allocate reply buffer of \a req.
1686 * \note After this, req->rq_repmsg is still not accessible.
1688 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1690 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1691 struct ptlrpc_sec_policy *policy;
1696 LASSERT(ctx->cc_sec);
1697 LASSERT(ctx->cc_sec->ps_policy);
1702 policy = ctx->cc_sec->ps_policy;
1703 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1707 * Used by ptlrpc client to free reply buffer of \a req. After this
1708 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1710 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1712 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1713 struct ptlrpc_sec_policy *policy;
1718 LASSERT(ctx->cc_sec);
1719 LASSERT(ctx->cc_sec->ps_policy);
1720 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1722 if (req->rq_repbuf == NULL)
1724 LASSERT(req->rq_repbuf_len);
1726 policy = ctx->cc_sec->ps_policy;
1727 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1728 req->rq_repmsg = NULL;
1731 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1733 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1734 struct ptlrpc_cli_ctx *ctx)
1736 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1738 if (!policy->sp_cops->install_rctx)
1740 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1743 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1744 struct ptlrpc_svc_ctx *ctx)
1746 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1748 if (!policy->sp_sops->install_rctx)
1750 return policy->sp_sops->install_rctx(imp, ctx);
1753 /* Get SELinux policy info from userspace */
1754 static int sepol_helper(struct obd_import *imp)
1756 char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1758 [0] = "/usr/sbin/l_getsepol",
1760 [2] = NULL, /* obd type */
1762 [4] = NULL, /* obd name */
1764 [6] = mtime_str, /* policy mtime */
1766 [8] = mode_str, /* enforcing mode */
1771 [1] = "PATH=/sbin:/usr/sbin",
1777 if (imp == NULL || imp->imp_obd == NULL ||
1778 imp->imp_obd->obd_type == NULL) {
1781 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1782 argv[4] = imp->imp_obd->obd_name;
1783 spin_lock(&imp->imp_sec->ps_lock);
1784 if (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1785 imp->imp_sec->ps_sepol[0] == '\0') {
1786 /* ps_sepol has not been initialized */
1792 mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1793 snprintf(mtime_str, sizeof(mtime_str), "%lld",
1794 mtime_ms / MSEC_PER_SEC);
1795 mode_str[0] = imp->imp_sec->ps_sepol[0];
1797 spin_unlock(&imp->imp_sec->ps_lock);
1798 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1805 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1809 if (send_sepol == 0 || !selinux_is_enabled())
1812 if (send_sepol == -1)
1813 /* send_sepol == -1 means fetch sepol status every time */
1816 spin_lock(&imp_sec->ps_lock);
1817 checknext = imp_sec->ps_sepol_checknext;
1818 spin_unlock(&imp_sec->ps_lock);
1820 /* next check is too far in time, please update */
1821 if (ktime_after(checknext,
1822 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1825 if (ktime_before(ktime_get(), checknext))
1826 /* too early to fetch sepol status */
1830 /* define new sepol_checknext time */
1831 spin_lock(&imp_sec->ps_lock);
1832 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1833 ktime_set(send_sepol, 0));
1834 spin_unlock(&imp_sec->ps_lock);
1839 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1841 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1846 (req->rq_sepol)[0] = '\0';
1848 #ifndef HAVE_SELINUX
1849 if (unlikely(send_sepol != 0))
1851 "Client cannot report SELinux status, it was not built against libselinux.\n");
1855 if (send_sepol == 0 || !selinux_is_enabled())
1858 if (imp_sec == NULL)
1861 /* Retrieve SELinux status info */
1862 if (sptlrpc_sepol_needs_check(imp_sec))
1863 rc = sepol_helper(req->rq_import);
1864 if (likely(rc == 0)) {
1865 spin_lock(&imp_sec->ps_lock);
1866 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1867 sizeof(req->rq_sepol));
1868 spin_unlock(&imp_sec->ps_lock);
1873 EXPORT_SYMBOL(sptlrpc_get_sepol);
1876 * server side security
1879 static int flavor_allowed(struct sptlrpc_flavor *exp,
1880 struct ptlrpc_request *req)
1882 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1884 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1887 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1888 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1889 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1890 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1896 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1899 * Given an export \a exp, check whether the flavor of incoming \a req
1900 * is allowed by the export \a exp. Main logic is about taking care of
1901 * changing configurations. Return 0 means success.
1903 int sptlrpc_target_export_check(struct obd_export *exp,
1904 struct ptlrpc_request *req)
1906 struct sptlrpc_flavor flavor;
1912 * client side export has no imp_reverse, skip
1913 * FIXME maybe we should check flavor this as well???
1915 if (exp->exp_imp_reverse == NULL)
1918 /* don't care about ctx fini rpc */
1919 if (req->rq_ctx_fini)
1922 spin_lock(&exp->exp_lock);
1925 * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1926 * the first req with the new flavor, then treat it as current flavor,
1927 * adapt reverse sec according to it.
1928 * note the first rpc with new flavor might not be with root ctx, in
1929 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1931 if (unlikely(exp->exp_flvr_changed) &&
1932 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1934 * make the new flavor as "current", and old ones as
1937 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1938 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1939 flavor = exp->exp_flvr_old[1];
1940 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1941 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1942 exp->exp_flvr_old[0] = exp->exp_flvr;
1943 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1944 EXP_FLVR_UPDATE_EXPIRE;
1945 exp->exp_flvr = flavor;
1947 /* flavor change finished */
1948 exp->exp_flvr_changed = 0;
1949 LASSERT(exp->exp_flvr_adapt == 1);
1951 /* if it's gss, we only interested in root ctx init */
1952 if (req->rq_auth_gss &&
1953 !(req->rq_ctx_init &&
1954 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1955 req->rq_auth_usr_ost))) {
1956 spin_unlock(&exp->exp_lock);
1957 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1958 req->rq_auth_gss, req->rq_ctx_init,
1959 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1960 req->rq_auth_usr_ost);
1964 exp->exp_flvr_adapt = 0;
1965 spin_unlock(&exp->exp_lock);
1967 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1968 req->rq_svc_ctx, &flavor);
1972 * if it equals to the current flavor, we accept it, but need to
1973 * dealing with reverse sec/ctx
1975 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1977 * most cases should return here, we only interested in
1980 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1981 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1982 !req->rq_auth_usr_ost)) {
1983 spin_unlock(&exp->exp_lock);
1988 * if flavor just changed, we should not proceed, just leave
1989 * it and current flavor will be discovered and replaced
1990 * shortly, and let _this_ rpc pass through
1992 if (exp->exp_flvr_changed) {
1993 LASSERT(exp->exp_flvr_adapt);
1994 spin_unlock(&exp->exp_lock);
1998 if (exp->exp_flvr_adapt) {
1999 exp->exp_flvr_adapt = 0;
2000 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2001 exp, exp->exp_flvr.sf_rpc,
2002 exp->exp_flvr_old[0].sf_rpc,
2003 exp->exp_flvr_old[1].sf_rpc);
2004 flavor = exp->exp_flvr;
2005 spin_unlock(&exp->exp_lock);
2007 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2012 "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2013 exp, exp->exp_flvr.sf_rpc,
2014 exp->exp_flvr_old[0].sf_rpc,
2015 exp->exp_flvr_old[1].sf_rpc);
2016 spin_unlock(&exp->exp_lock);
2018 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2023 if (exp->exp_flvr_expire[0]) {
2024 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2025 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2027 "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2028 exp, exp->exp_flvr.sf_rpc,
2029 exp->exp_flvr_old[0].sf_rpc,
2030 exp->exp_flvr_old[1].sf_rpc,
2031 (s64)(exp->exp_flvr_expire[0] -
2032 ktime_get_real_seconds()));
2033 spin_unlock(&exp->exp_lock);
2037 CDEBUG(D_SEC, "mark middle expired\n");
2038 exp->exp_flvr_expire[0] = 0;
2040 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2041 exp->exp_flvr.sf_rpc,
2042 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2043 req->rq_flvr.sf_rpc);
2047 * now it doesn't match the current flavor, the only chance we can
2048 * accept it is match the old flavors which is not expired.
2050 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2051 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2052 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2053 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2055 exp->exp_flvr.sf_rpc,
2056 exp->exp_flvr_old[0].sf_rpc,
2057 exp->exp_flvr_old[1].sf_rpc,
2058 (s64)(exp->exp_flvr_expire[1] -
2059 ktime_get_real_seconds()));
2060 spin_unlock(&exp->exp_lock);
2064 CDEBUG(D_SEC, "mark oldest expired\n");
2065 exp->exp_flvr_expire[1] = 0;
2067 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2068 exp, exp->exp_flvr.sf_rpc,
2069 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2070 req->rq_flvr.sf_rpc);
2072 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2073 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2074 exp->exp_flvr_old[1].sf_rpc);
2077 spin_unlock(&exp->exp_lock);
2079 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2080 exp, exp->exp_obd->obd_name,
2081 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2082 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2083 req->rq_flvr.sf_rpc,
2084 exp->exp_flvr.sf_rpc,
2085 exp->exp_flvr_old[0].sf_rpc,
2086 exp->exp_flvr_expire[0] ?
2087 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2088 exp->exp_flvr_old[1].sf_rpc,
2089 exp->exp_flvr_expire[1] ?
2090 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2093 EXPORT_SYMBOL(sptlrpc_target_export_check);
2095 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2096 struct sptlrpc_rule_set *rset)
2098 struct obd_export *exp;
2099 struct sptlrpc_flavor new_flvr;
2103 spin_lock(&obd->obd_dev_lock);
2105 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2106 if (exp->exp_connection == NULL)
2110 * note if this export had just been updated flavor
2111 * (exp_flvr_changed == 1), this will override the
2114 spin_lock(&exp->exp_lock);
2115 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2116 exp->exp_connection->c_peer.nid,
2118 if (exp->exp_flvr_changed ||
2119 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2120 exp->exp_flvr_old[1] = new_flvr;
2121 exp->exp_flvr_expire[1] = 0;
2122 exp->exp_flvr_changed = 1;
2123 exp->exp_flvr_adapt = 1;
2125 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2126 exp, sptlrpc_part2name(exp->exp_sp_peer),
2127 exp->exp_flvr.sf_rpc,
2128 exp->exp_flvr_old[1].sf_rpc);
2130 spin_unlock(&exp->exp_lock);
2133 spin_unlock(&obd->obd_dev_lock);
2135 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2137 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2139 /* peer's claim is unreliable unless gss is being used */
2140 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2143 switch (req->rq_sp_from) {
2145 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2146 /* The below message is checked in sanity-sec test_33 */
2147 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2148 svc_rc = SECSVC_DROP;
2152 if (!req->rq_auth_usr_mdt) {
2153 /* The below message is checked in sanity-sec test_33 */
2154 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2155 svc_rc = SECSVC_DROP;
2159 if (!req->rq_auth_usr_ost) {
2160 /* The below message is checked in sanity-sec test_33 */
2161 DEBUG_REQ(D_ERROR, req, "faked source OST");
2162 svc_rc = SECSVC_DROP;
2167 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2168 !req->rq_auth_usr_ost) {
2169 /* The below message is checked in sanity-sec test_33 */
2170 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2171 svc_rc = SECSVC_DROP;
2176 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2177 svc_rc = SECSVC_DROP;
2184 * Used by ptlrpc server, to perform transformation upon request message of
2185 * incoming \a req. This must be the first thing to do with an incoming
2186 * request in ptlrpc layer.
2188 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2189 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2190 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2191 * reply message has been prepared.
2192 * \retval SECSVC_DROP failed, this request should be dropped.
2194 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2196 struct ptlrpc_sec_policy *policy;
2197 struct lustre_msg *msg = req->rq_reqbuf;
2203 LASSERT(req->rq_reqmsg == NULL);
2204 LASSERT(req->rq_repmsg == NULL);
2205 LASSERT(req->rq_svc_ctx == NULL);
2207 req->rq_req_swab_mask = 0;
2209 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2212 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2216 CERROR("error unpacking request from %s x%llu\n",
2217 libcfs_id2str(req->rq_peer), req->rq_xid);
2218 RETURN(SECSVC_DROP);
2221 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2222 req->rq_sp_from = LUSTRE_SP_ANY;
2223 req->rq_auth_uid = -1; /* set to INVALID_UID */
2224 req->rq_auth_mapped_uid = -1;
2226 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2228 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2229 RETURN(SECSVC_DROP);
2232 LASSERT(policy->sp_sops->accept);
2233 rc = policy->sp_sops->accept(req);
2234 sptlrpc_policy_put(policy);
2235 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2236 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2239 * if it's not null flavor (which means embedded packing msg),
2240 * reset the swab mask for the comming inner msg unpacking.
2242 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2243 req->rq_req_swab_mask = 0;
2245 /* sanity check for the request source */
2246 rc = sptlrpc_svc_check_from(req, rc);
2251 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2252 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2253 * a buffer of \a msglen size.
2255 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2257 struct ptlrpc_sec_policy *policy;
2258 struct ptlrpc_reply_state *rs;
2263 LASSERT(req->rq_svc_ctx);
2264 LASSERT(req->rq_svc_ctx->sc_policy);
2266 policy = req->rq_svc_ctx->sc_policy;
2267 LASSERT(policy->sp_sops->alloc_rs);
2269 rc = policy->sp_sops->alloc_rs(req, msglen);
2270 if (unlikely(rc == -ENOMEM)) {
2271 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2273 if (svcpt->scp_service->srv_max_reply_size <
2274 msglen + sizeof(struct ptlrpc_reply_state)) {
2275 /* Just return failure if the size is too big */
2276 CERROR("size of message is too big (%zd), %d allowed\n",
2277 msglen + sizeof(struct ptlrpc_reply_state),
2278 svcpt->scp_service->srv_max_reply_size);
2282 /* failed alloc, try emergency pool */
2283 rs = lustre_get_emerg_rs(svcpt);
2287 req->rq_reply_state = rs;
2288 rc = policy->sp_sops->alloc_rs(req, msglen);
2290 lustre_put_emerg_rs(rs);
2291 req->rq_reply_state = NULL;
2296 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2302 * Used by ptlrpc server, to perform transformation upon reply message.
2304 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2305 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2307 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2309 struct ptlrpc_sec_policy *policy;
2314 LASSERT(req->rq_svc_ctx);
2315 LASSERT(req->rq_svc_ctx->sc_policy);
2317 policy = req->rq_svc_ctx->sc_policy;
2318 LASSERT(policy->sp_sops->authorize);
2320 rc = policy->sp_sops->authorize(req);
2321 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2327 * Used by ptlrpc server, to free reply_state.
2329 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2331 struct ptlrpc_sec_policy *policy;
2332 unsigned int prealloc;
2336 LASSERT(rs->rs_svc_ctx);
2337 LASSERT(rs->rs_svc_ctx->sc_policy);
2339 policy = rs->rs_svc_ctx->sc_policy;
2340 LASSERT(policy->sp_sops->free_rs);
2342 prealloc = rs->rs_prealloc;
2343 policy->sp_sops->free_rs(rs);
2346 lustre_put_emerg_rs(rs);
2350 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2352 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2355 atomic_inc(&ctx->sc_refcount);
2358 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2360 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2365 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2366 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2367 if (ctx->sc_policy->sp_sops->free_ctx)
2368 ctx->sc_policy->sp_sops->free_ctx(ctx);
2370 req->rq_svc_ctx = NULL;
2373 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2375 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2380 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2381 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2382 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2384 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2391 * Perform transformation upon bulk data pointed by \a desc. This is called
2392 * before transforming the request message.
2394 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2395 struct ptlrpc_bulk_desc *desc)
2397 struct ptlrpc_cli_ctx *ctx;
2399 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2401 if (!req->rq_pack_bulk)
2404 ctx = req->rq_cli_ctx;
2405 if (ctx->cc_ops->wrap_bulk)
2406 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2409 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2412 * This is called after unwrap the reply message.
2413 * return nob of actual plain text size received, or error code.
2415 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2416 struct ptlrpc_bulk_desc *desc,
2419 struct ptlrpc_cli_ctx *ctx;
2422 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2424 if (!req->rq_pack_bulk)
2425 return desc->bd_nob_transferred;
2427 ctx = req->rq_cli_ctx;
2428 if (ctx->cc_ops->unwrap_bulk) {
2429 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2433 return desc->bd_nob_transferred;
2435 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2438 * This is called after unwrap the reply message.
2439 * return 0 for success or error code.
2441 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2442 struct ptlrpc_bulk_desc *desc)
2444 struct ptlrpc_cli_ctx *ctx;
2447 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2449 if (!req->rq_pack_bulk)
2452 ctx = req->rq_cli_ctx;
2453 if (ctx->cc_ops->unwrap_bulk) {
2454 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2460 * if everything is going right, nob should equals to nob_transferred.
2461 * in case of privacy mode, nob_transferred needs to be adjusted.
2463 if (desc->bd_nob != desc->bd_nob_transferred) {
2464 CERROR("nob %d doesn't match transferred nob %d\n",
2465 desc->bd_nob, desc->bd_nob_transferred);
2471 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2473 #ifdef HAVE_SERVER_SUPPORT
2475 * Performe transformation upon outgoing bulk read.
2477 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2478 struct ptlrpc_bulk_desc *desc)
2480 struct ptlrpc_svc_ctx *ctx;
2482 LASSERT(req->rq_bulk_read);
2484 if (!req->rq_pack_bulk)
2487 ctx = req->rq_svc_ctx;
2488 if (ctx->sc_policy->sp_sops->wrap_bulk)
2489 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2493 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2496 * Performe transformation upon incoming bulk write.
2498 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2499 struct ptlrpc_bulk_desc *desc)
2501 struct ptlrpc_svc_ctx *ctx;
2504 LASSERT(req->rq_bulk_write);
2507 * if it's in privacy mode, transferred should >= expected; otherwise
2508 * transferred should == expected.
2510 if (desc->bd_nob_transferred < desc->bd_nob ||
2511 (desc->bd_nob_transferred > desc->bd_nob &&
2512 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2513 SPTLRPC_BULK_SVC_PRIV)) {
2514 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2515 desc->bd_nob_transferred, desc->bd_nob);
2519 if (!req->rq_pack_bulk)
2522 ctx = req->rq_svc_ctx;
2523 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2524 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2526 CERROR("error unwrap bulk: %d\n", rc);
2529 /* return 0 to allow reply be sent */
2532 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2535 * Prepare buffers for incoming bulk write.
2537 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2538 struct ptlrpc_bulk_desc *desc)
2540 struct ptlrpc_svc_ctx *ctx;
2542 LASSERT(req->rq_bulk_write);
2544 if (!req->rq_pack_bulk)
2547 ctx = req->rq_svc_ctx;
2548 if (ctx->sc_policy->sp_sops->prep_bulk)
2549 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2553 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2555 #endif /* HAVE_SERVER_SUPPORT */
2558 * user descriptor helpers
2561 int sptlrpc_current_user_desc_size(void)
2565 ngroups = current_cred()->group_info->ngroups;
2567 if (ngroups > LUSTRE_MAX_GROUPS)
2568 ngroups = LUSTRE_MAX_GROUPS;
2569 return sptlrpc_user_desc_size(ngroups);
2571 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2573 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2575 struct ptlrpc_user_desc *pud;
2578 pud = lustre_msg_buf(msg, offset, 0);
2580 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2581 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2582 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2583 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2584 pud->pud_cap = cfs_curproc_cap_pack();
2585 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2588 ngroups = current_cred()->group_info->ngroups;
2589 if (pud->pud_ngroups > ngroups)
2590 pud->pud_ngroups = ngroups;
2591 #ifdef HAVE_GROUP_INFO_GID
2592 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2593 pud->pud_ngroups * sizeof(__u32));
2594 #else /* !HAVE_GROUP_INFO_GID */
2595 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2596 pud->pud_ngroups * sizeof(__u32));
2597 #endif /* HAVE_GROUP_INFO_GID */
2598 task_unlock(current);
2602 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2604 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2606 struct ptlrpc_user_desc *pud;
2609 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2614 __swab32s(&pud->pud_uid);
2615 __swab32s(&pud->pud_gid);
2616 __swab32s(&pud->pud_fsuid);
2617 __swab32s(&pud->pud_fsgid);
2618 __swab32s(&pud->pud_cap);
2619 __swab32s(&pud->pud_ngroups);
2622 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2623 CERROR("%u groups is too large\n", pud->pud_ngroups);
2627 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2628 msg->lm_buflens[offset]) {
2629 CERROR("%u groups are claimed but bufsize only %u\n",
2630 pud->pud_ngroups, msg->lm_buflens[offset]);
2635 for (i = 0; i < pud->pud_ngroups; i++)
2636 __swab32s(&pud->pud_groups[i]);
2641 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2647 const char *sec2target_str(struct ptlrpc_sec *sec)
2649 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2651 if (sec_is_reverse(sec))
2653 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2655 EXPORT_SYMBOL(sec2target_str);
2658 * return true if the bulk data is protected
2660 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2662 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2663 case SPTLRPC_BULK_SVC_INTG:
2664 case SPTLRPC_BULK_SVC_PRIV:
2670 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2673 * crypto API helper/alloc blkciper
2677 * initialize/finalize
2680 int sptlrpc_init(void)
2684 rwlock_init(&policy_lock);
2686 rc = sptlrpc_gc_init();
2690 rc = sptlrpc_conf_init();
2694 rc = sptlrpc_enc_pool_init();
2698 rc = sptlrpc_null_init();
2702 rc = sptlrpc_plain_init();
2706 rc = sptlrpc_lproc_init();
2713 sptlrpc_plain_fini();
2715 sptlrpc_null_fini();
2717 sptlrpc_enc_pool_fini();
2719 sptlrpc_conf_fini();
2726 void sptlrpc_fini(void)
2728 sptlrpc_lproc_fini();
2729 sptlrpc_plain_fini();
2730 sptlrpc_null_fini();
2731 sptlrpc_enc_pool_fini();
2732 sptlrpc_conf_fini();