1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
40 * Author: Eric Mei <ericm@clusterfs.com>
46 #define DEBUG_SUBSYSTEM S_SEC
48 #include <libcfs/libcfs.h>
50 #include <liblustre.h>
51 #include <libcfs/list.h>
53 #include <linux/crypto.h>
54 #include <linux/key.h>
58 #include <obd_class.h>
59 #include <obd_support.h>
60 #include <lustre_net.h>
61 #include <lustre_import.h>
62 #include <lustre_dlm.h>
63 #include <lustre_sec.h>
65 #include "ptlrpc_internal.h"
67 /***********************************************
69 ***********************************************/
71 static cfs_rwlock_t policy_lock;
72 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
76 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
78 __u16 number = policy->sp_policy;
80 LASSERT(policy->sp_name);
81 LASSERT(policy->sp_cops);
82 LASSERT(policy->sp_sops);
84 if (number >= SPTLRPC_POLICY_MAX)
87 cfs_write_lock(&policy_lock);
88 if (unlikely(policies[number])) {
89 cfs_write_unlock(&policy_lock);
92 policies[number] = policy;
93 cfs_write_unlock(&policy_lock);
95 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
98 EXPORT_SYMBOL(sptlrpc_register_policy);
100 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
102 __u16 number = policy->sp_policy;
104 LASSERT(number < SPTLRPC_POLICY_MAX);
106 cfs_write_lock(&policy_lock);
107 if (unlikely(policies[number] == NULL)) {
108 cfs_write_unlock(&policy_lock);
109 CERROR("%s: already unregistered\n", policy->sp_name);
113 LASSERT(policies[number] == policy);
114 policies[number] = NULL;
115 cfs_write_unlock(&policy_lock);
117 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
120 EXPORT_SYMBOL(sptlrpc_unregister_policy);
123 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
125 static CFS_DEFINE_MUTEX(load_mutex);
126 static cfs_atomic_t loaded = CFS_ATOMIC_INIT(0);
127 struct ptlrpc_sec_policy *policy;
128 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
131 if (number >= SPTLRPC_POLICY_MAX)
135 cfs_read_lock(&policy_lock);
136 policy = policies[number];
137 if (policy && !cfs_try_module_get(policy->sp_owner))
140 flag = cfs_atomic_read(&loaded);
141 cfs_read_unlock(&policy_lock);
143 if (policy != NULL || flag != 0 ||
144 number != SPTLRPC_POLICY_GSS)
147 /* try to load gss module, once */
148 cfs_mutex_lock(&load_mutex);
149 if (cfs_atomic_read(&loaded) == 0) {
150 if (cfs_request_module("ptlrpc_gss") == 0)
152 "module ptlrpc_gss loaded on demand\n");
154 CERROR("Unable to load module ptlrpc_gss\n");
156 cfs_atomic_set(&loaded, 1);
158 cfs_mutex_unlock(&load_mutex);
164 __u32 sptlrpc_name2flavor_base(const char *name)
166 if (!strcmp(name, "null"))
167 return SPTLRPC_FLVR_NULL;
168 if (!strcmp(name, "plain"))
169 return SPTLRPC_FLVR_PLAIN;
170 if (!strcmp(name, "krb5n"))
171 return SPTLRPC_FLVR_KRB5N;
172 if (!strcmp(name, "krb5a"))
173 return SPTLRPC_FLVR_KRB5A;
174 if (!strcmp(name, "krb5i"))
175 return SPTLRPC_FLVR_KRB5I;
176 if (!strcmp(name, "krb5p"))
177 return SPTLRPC_FLVR_KRB5P;
179 return SPTLRPC_FLVR_INVALID;
181 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
183 const char *sptlrpc_flavor2name_base(__u32 flvr)
185 __u32 base = SPTLRPC_FLVR_BASE(flvr);
187 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
189 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
191 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
200 CERROR("invalid wire flavor 0x%x\n", flvr);
203 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
205 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
206 char *buf, int bufsize)
208 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
209 snprintf(buf, bufsize, "hash:%s",
210 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
212 snprintf(buf, bufsize, "%s",
213 sptlrpc_flavor2name_base(sf->sf_rpc));
215 buf[bufsize - 1] = '\0';
218 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
220 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
222 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
225 * currently we don't support customized bulk specification for
226 * flavors other than plain
228 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
232 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
233 strncat(buf, bspec, bufsize);
236 buf[bufsize - 1] = '\0';
239 EXPORT_SYMBOL(sptlrpc_flavor2name);
241 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
245 if (flags & PTLRPC_SEC_FL_REVERSE)
246 strncat(buf, "reverse,", bufsize);
247 if (flags & PTLRPC_SEC_FL_ROOTONLY)
248 strncat(buf, "rootonly,", bufsize);
249 if (flags & PTLRPC_SEC_FL_UDESC)
250 strncat(buf, "udesc,", bufsize);
251 if (flags & PTLRPC_SEC_FL_BULK)
252 strncat(buf, "bulk,", bufsize);
254 strncat(buf, "-,", bufsize);
256 buf[bufsize - 1] = '\0';
259 EXPORT_SYMBOL(sptlrpc_secflags2str);
261 /**************************************************
262 * client context APIs *
263 **************************************************/
266 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
268 struct vfs_cred vcred;
269 int create = 1, remove_dead = 1;
272 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
274 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
275 PTLRPC_SEC_FL_ROOTONLY)) {
278 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
283 vcred.vc_uid = cfs_curproc_uid();
284 vcred.vc_gid = cfs_curproc_gid();
287 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
288 create, remove_dead);
291 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
293 cfs_atomic_inc(&ctx->cc_refcount);
296 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
298 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
300 struct ptlrpc_sec *sec = ctx->cc_sec;
303 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
305 if (!cfs_atomic_dec_and_test(&ctx->cc_refcount))
308 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
310 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
313 * Expire the client context immediately.
315 * \pre Caller must hold at least 1 reference on the \a ctx.
317 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
319 LASSERT(ctx->cc_ops->die);
320 ctx->cc_ops->die(ctx, 0);
322 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
325 * To wake up the threads who are waiting for this client context. Called
326 * after some status change happened on \a ctx.
328 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
330 struct ptlrpc_request *req, *next;
332 cfs_spin_lock(&ctx->cc_lock);
333 cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list,
335 cfs_list_del_init(&req->rq_ctx_chain);
336 ptlrpc_client_wake_req(req);
338 cfs_spin_unlock(&ctx->cc_lock);
340 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
342 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
344 LASSERT(ctx->cc_ops);
346 if (ctx->cc_ops->display == NULL)
349 return ctx->cc_ops->display(ctx, buf, bufsize);
352 static int import_sec_check_expire(struct obd_import *imp)
356 cfs_spin_lock(&imp->imp_lock);
357 if (imp->imp_sec_expire &&
358 imp->imp_sec_expire < cfs_time_current_sec()) {
360 imp->imp_sec_expire = 0;
362 cfs_spin_unlock(&imp->imp_lock);
367 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
368 return sptlrpc_import_sec_adapt(imp, NULL, 0);
371 static int import_sec_validate_get(struct obd_import *imp,
372 struct ptlrpc_sec **sec)
376 if (unlikely(imp->imp_sec_expire)) {
377 rc = import_sec_check_expire(imp);
382 *sec = sptlrpc_import_sec_ref(imp);
384 CERROR("import %p (%s) with no sec\n",
385 imp, ptlrpc_import_state_name(imp->imp_state));
389 if (unlikely((*sec)->ps_dying)) {
390 CERROR("attempt to use dying sec %p\n", sec);
391 sptlrpc_sec_put(*sec);
399 * Given a \a req, find or allocate a appropriate context for it.
400 * \pre req->rq_cli_ctx == NULL.
402 * \retval 0 succeed, and req->rq_cli_ctx is set.
403 * \retval -ev error number, and req->rq_cli_ctx == NULL.
405 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
407 struct obd_import *imp = req->rq_import;
408 struct ptlrpc_sec *sec;
412 LASSERT(!req->rq_cli_ctx);
415 rc = import_sec_validate_get(imp, &sec);
419 req->rq_cli_ctx = get_my_ctx(sec);
421 sptlrpc_sec_put(sec);
423 if (!req->rq_cli_ctx) {
424 CERROR("req %p: fail to get context\n", req);
432 * Drop the context for \a req.
433 * \pre req->rq_cli_ctx != NULL.
434 * \post req->rq_cli_ctx == NULL.
436 * If \a sync == 0, this function should return quickly without sleep;
437 * otherwise it might trigger and wait for the whole process of sending
438 * an context-destroying rpc to server.
440 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
445 LASSERT(req->rq_cli_ctx);
447 /* request might be asked to release earlier while still
448 * in the context waiting list.
450 if (!cfs_list_empty(&req->rq_ctx_chain)) {
451 cfs_spin_lock(&req->rq_cli_ctx->cc_lock);
452 cfs_list_del_init(&req->rq_ctx_chain);
453 cfs_spin_unlock(&req->rq_cli_ctx->cc_lock);
456 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
457 req->rq_cli_ctx = NULL;
462 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
463 struct ptlrpc_cli_ctx *oldctx,
464 struct ptlrpc_cli_ctx *newctx)
466 struct sptlrpc_flavor old_flvr;
467 char *reqmsg = NULL; /* to workaround old gcc */
471 LASSERT(req->rq_reqmsg);
472 LASSERT(req->rq_reqlen);
473 LASSERT(req->rq_replen);
475 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
476 "switch sec %p(%s) -> %p(%s)\n", req,
477 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
478 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
479 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
480 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
483 old_flvr = req->rq_flvr;
485 /* save request message */
486 reqmsg_size = req->rq_reqlen;
487 if (reqmsg_size != 0) {
488 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
491 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
494 /* release old req/rep buf */
495 req->rq_cli_ctx = oldctx;
496 sptlrpc_cli_free_reqbuf(req);
497 sptlrpc_cli_free_repbuf(req);
498 req->rq_cli_ctx = newctx;
500 /* recalculate the flavor */
501 sptlrpc_req_set_flavor(req, 0);
503 /* alloc new request buffer
504 * we don't need to alloc reply buffer here, leave it to the
505 * rest procedure of ptlrpc */
506 if (reqmsg_size != 0) {
507 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
509 LASSERT(req->rq_reqmsg);
510 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
512 CWARN("failed to alloc reqbuf: %d\n", rc);
513 req->rq_flvr = old_flvr;
516 OBD_FREE_LARGE(reqmsg, reqmsg_size);
522 * If current context of \a req is dead somehow, e.g. we just switched flavor
523 * thus marked original contexts dead, we'll find a new context for it. if
524 * no switch is needed, \a req will end up with the same context.
526 * \note a request must have a context, to keep other parts of code happy.
527 * In any case of failure during the switching, we must restore the old one.
529 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
531 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
532 struct ptlrpc_cli_ctx *newctx;
538 sptlrpc_cli_ctx_get(oldctx);
539 sptlrpc_req_put_ctx(req, 0);
541 rc = sptlrpc_req_get_ctx(req);
543 LASSERT(!req->rq_cli_ctx);
545 /* restore old ctx */
546 req->rq_cli_ctx = oldctx;
550 newctx = req->rq_cli_ctx;
553 if (unlikely(newctx == oldctx &&
554 cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
556 * still get the old dead ctx, usually means system too busy
559 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
560 newctx, newctx->cc_flags);
562 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
566 * it's possible newctx == oldctx if we're switching
567 * subflavor with the same sec.
569 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
571 /* restore old ctx */
572 sptlrpc_req_put_ctx(req, 0);
573 req->rq_cli_ctx = oldctx;
577 LASSERT(req->rq_cli_ctx == newctx);
580 sptlrpc_cli_ctx_put(oldctx, 1);
583 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
586 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
588 if (cli_ctx_is_refreshed(ctx))
594 int ctx_refresh_timeout(void *data)
596 struct ptlrpc_request *req = data;
599 /* conn_cnt is needed in expire_one_request */
600 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
602 rc = ptlrpc_expire_one_request(req, 1);
603 /* if we started recovery, we should mark this ctx dead; otherwise
604 * in case of lgssd died nobody would retire this ctx, following
605 * connecting will still find the same ctx thus cause deadlock.
606 * there's an assumption that expire time of the request should be
607 * later than the context refresh expire time.
610 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
615 void ctx_refresh_interrupt(void *data)
617 struct ptlrpc_request *req = data;
619 cfs_spin_lock(&req->rq_lock);
621 cfs_spin_unlock(&req->rq_lock);
625 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
627 cfs_spin_lock(&ctx->cc_lock);
628 if (!cfs_list_empty(&req->rq_ctx_chain))
629 cfs_list_del_init(&req->rq_ctx_chain);
630 cfs_spin_unlock(&ctx->cc_lock);
634 * To refresh the context of \req, if it's not up-to-date.
637 * - = 0: wait until success or fatal error occur
638 * - > 0: timeout value (in seconds)
640 * The status of the context could be subject to be changed by other threads
641 * at any time. We allow this race, but once we return with 0, the caller will
642 * suppose it's uptodated and keep using it until the owning rpc is done.
644 * \retval 0 only if the context is uptodated.
645 * \retval -ev error number.
647 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
649 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
650 struct ptlrpc_sec *sec;
651 struct l_wait_info lwi;
657 if (req->rq_ctx_init || req->rq_ctx_fini)
661 * during the process a request's context might change type even
662 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
666 rc = import_sec_validate_get(req->rq_import, &sec);
670 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
671 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
672 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
673 req_off_ctx_list(req, ctx);
674 sptlrpc_req_replace_dead_ctx(req);
675 ctx = req->rq_cli_ctx;
677 sptlrpc_sec_put(sec);
679 if (cli_ctx_is_eternal(ctx))
682 if (unlikely(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
683 LASSERT(ctx->cc_ops->refresh);
684 ctx->cc_ops->refresh(ctx);
686 LASSERT(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
688 LASSERT(ctx->cc_ops->validate);
689 if (ctx->cc_ops->validate(ctx) == 0) {
690 req_off_ctx_list(req, ctx);
694 if (unlikely(cfs_test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
695 cfs_spin_lock(&req->rq_lock);
697 cfs_spin_unlock(&req->rq_lock);
698 req_off_ctx_list(req, ctx);
703 * There's a subtle issue for resending RPCs, suppose following
705 * 1. the request was sent to server.
706 * 2. recovery was kicked start, after finished the request was
708 * 3. resend the request.
709 * 4. old reply from server received, we accept and verify the reply.
710 * this has to be success, otherwise the error will be aware
712 * 5. new reply from server received, dropped by LNet.
714 * Note the xid of old & new request is the same. We can't simply
715 * change xid for the resent request because the server replies on
716 * it for reply reconstruction.
718 * Commonly the original context should be uptodate because we
719 * have a expiry nice time; server will keep its context because
720 * we at least hold a ref of old context which prevent context
721 * destroying RPC being sent. So server still can accept the request
722 * and finish the RPC. But if that's not the case:
723 * 1. If server side context has been trimmed, a NO_CONTEXT will
724 * be returned, gss_cli_ctx_verify/unseal will switch to new
726 * 2. Current context never be refreshed, then we are fine: we
727 * never really send request with old context before.
729 if (cfs_test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
730 unlikely(req->rq_reqmsg) &&
731 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
732 req_off_ctx_list(req, ctx);
736 if (unlikely(cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
737 req_off_ctx_list(req, ctx);
739 * don't switch ctx if import was deactivated
741 if (req->rq_import->imp_deactive) {
742 cfs_spin_lock(&req->rq_lock);
744 cfs_spin_unlock(&req->rq_lock);
748 rc = sptlrpc_req_replace_dead_ctx(req);
750 LASSERT(ctx == req->rq_cli_ctx);
751 CERROR("req %p: failed to replace dead ctx %p: %d\n",
753 cfs_spin_lock(&req->rq_lock);
755 cfs_spin_unlock(&req->rq_lock);
759 ctx = req->rq_cli_ctx;
764 * Now we're sure this context is during upcall, add myself into
767 cfs_spin_lock(&ctx->cc_lock);
768 if (cfs_list_empty(&req->rq_ctx_chain))
769 cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
770 cfs_spin_unlock(&ctx->cc_lock);
773 RETURN(-EWOULDBLOCK);
775 /* Clear any flags that may be present from previous sends */
776 LASSERT(req->rq_receiving_reply == 0);
777 cfs_spin_lock(&req->rq_lock);
779 req->rq_timedout = 0;
782 cfs_spin_unlock(&req->rq_lock);
784 lwi = LWI_TIMEOUT_INTR(timeout * CFS_HZ, ctx_refresh_timeout,
785 ctx_refresh_interrupt, req);
786 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
789 * following cases could lead us here:
790 * - successfully refreshed;
792 * - timedout, and we don't want recover from the failure;
793 * - timedout, and waked up upon recovery finished;
794 * - someone else mark this ctx dead by force;
795 * - someone invalidate the req and call ptlrpc_client_wake_req(),
796 * e.g. ptlrpc_abort_inflight();
798 if (!cli_ctx_is_refreshed(ctx)) {
799 /* timed out or interruptted */
800 req_off_ctx_list(req, ctx);
810 * Initialize flavor settings for \a req, according to \a opcode.
812 * \note this could be called in two situations:
813 * - new request from ptlrpc_pre_req(), with proper @opcode
814 * - old request which changed ctx in the middle, with @opcode == 0
816 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
818 struct ptlrpc_sec *sec;
820 LASSERT(req->rq_import);
821 LASSERT(req->rq_cli_ctx);
822 LASSERT(req->rq_cli_ctx->cc_sec);
823 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
825 /* special security flags accoding to opcode */
829 case MGS_CONFIG_READ:
830 req->rq_bulk_read = 1;
834 req->rq_bulk_write = 1;
837 req->rq_ctx_init = 1;
840 req->rq_ctx_fini = 1;
843 /* init/fini rpc won't be resend, so can't be here */
844 LASSERT(req->rq_ctx_init == 0);
845 LASSERT(req->rq_ctx_fini == 0);
847 /* cleanup flags, which should be recalculated */
848 req->rq_pack_udesc = 0;
849 req->rq_pack_bulk = 0;
853 sec = req->rq_cli_ctx->cc_sec;
855 cfs_spin_lock(&sec->ps_lock);
856 req->rq_flvr = sec->ps_flvr;
857 cfs_spin_unlock(&sec->ps_lock);
859 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
861 if (unlikely(req->rq_ctx_init))
862 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
863 else if (unlikely(req->rq_ctx_fini))
864 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
866 /* user descriptor flag, null security can't do it anyway */
867 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
868 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
869 req->rq_pack_udesc = 1;
871 /* bulk security flag */
872 if ((req->rq_bulk_read || req->rq_bulk_write) &&
873 sptlrpc_flavor_has_bulk(&req->rq_flvr))
874 req->rq_pack_bulk = 1;
877 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
879 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
882 LASSERT(req->rq_clrbuf);
883 if (req->rq_pool || !req->rq_reqbuf)
886 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
887 req->rq_reqbuf = NULL;
888 req->rq_reqbuf_len = 0;
892 * Given an import \a imp, check whether current user has a valid context
893 * or not. We may create a new context and try to refresh it, and try
894 * repeatedly try in case of non-fatal errors. Return 0 means success.
896 int sptlrpc_import_check_ctx(struct obd_import *imp)
898 struct ptlrpc_sec *sec;
899 struct ptlrpc_cli_ctx *ctx;
900 struct ptlrpc_request *req = NULL;
906 sec = sptlrpc_import_sec_ref(imp);
907 ctx = get_my_ctx(sec);
908 sptlrpc_sec_put(sec);
913 if (cli_ctx_is_eternal(ctx) ||
914 ctx->cc_ops->validate(ctx) == 0) {
915 sptlrpc_cli_ctx_put(ctx, 1);
919 if (cli_ctx_is_error(ctx)) {
920 sptlrpc_cli_ctx_put(ctx, 1);
928 cfs_spin_lock_init(&req->rq_lock);
929 cfs_atomic_set(&req->rq_refcount, 10000);
930 CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
931 cfs_waitq_init(&req->rq_reply_waitq);
932 cfs_waitq_init(&req->rq_set_waitq);
933 req->rq_import = imp;
934 req->rq_flvr = sec->ps_flvr;
935 req->rq_cli_ctx = ctx;
937 rc = sptlrpc_req_refresh_ctx(req, 0);
938 LASSERT(cfs_list_empty(&req->rq_ctx_chain));
939 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
946 * Used by ptlrpc client, to perform the pre-defined security transformation
947 * upon the request message of \a req. After this function called,
948 * req->rq_reqmsg is still accessible as clear text.
950 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
952 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
957 LASSERT(ctx->cc_sec);
958 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
960 /* we wrap bulk request here because now we can be sure
961 * the context is uptodate.
964 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
969 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
970 case SPTLRPC_SVC_NULL:
971 case SPTLRPC_SVC_AUTH:
972 case SPTLRPC_SVC_INTG:
973 LASSERT(ctx->cc_ops->sign);
974 rc = ctx->cc_ops->sign(ctx, req);
976 case SPTLRPC_SVC_PRIV:
977 LASSERT(ctx->cc_ops->seal);
978 rc = ctx->cc_ops->seal(ctx, req);
985 LASSERT(req->rq_reqdata_len);
986 LASSERT(req->rq_reqdata_len % 8 == 0);
987 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
993 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
995 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1000 LASSERT(ctx->cc_sec);
1001 LASSERT(req->rq_repbuf);
1002 LASSERT(req->rq_repdata);
1003 LASSERT(req->rq_repmsg == NULL);
1005 req->rq_rep_swab_mask = 0;
1007 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1010 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1014 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1018 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1019 CERROR("replied data length %d too small\n",
1020 req->rq_repdata_len);
1024 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1025 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1026 CERROR("reply policy %u doesn't match request policy %u\n",
1027 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1028 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1032 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1033 case SPTLRPC_SVC_NULL:
1034 case SPTLRPC_SVC_AUTH:
1035 case SPTLRPC_SVC_INTG:
1036 LASSERT(ctx->cc_ops->verify);
1037 rc = ctx->cc_ops->verify(ctx, req);
1039 case SPTLRPC_SVC_PRIV:
1040 LASSERT(ctx->cc_ops->unseal);
1041 rc = ctx->cc_ops->unseal(ctx, req);
1046 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1048 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1050 req->rq_rep_swab_mask = 0;
1055 * Used by ptlrpc client, to perform security transformation upon the reply
1056 * message of \a req. After return successfully, req->rq_repmsg points to
1057 * the reply message in clear text.
1059 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1062 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1064 LASSERT(req->rq_repbuf);
1065 LASSERT(req->rq_repdata == NULL);
1066 LASSERT(req->rq_repmsg == NULL);
1067 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1069 if (req->rq_reply_off == 0 &&
1070 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1071 CERROR("real reply with offset 0\n");
1075 if (req->rq_reply_off % 8 != 0) {
1076 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1080 req->rq_repdata = (struct lustre_msg *)
1081 (req->rq_repbuf + req->rq_reply_off);
1082 req->rq_repdata_len = req->rq_nob_received;
1084 return do_cli_unwrap_reply(req);
1088 * Used by ptlrpc client, to perform security transformation upon the early
1089 * reply message of \a req. We expect the rq_reply_off is 0, and
1090 * rq_nob_received is the early reply size.
1092 * Because the receive buffer might be still posted, the reply data might be
1093 * changed at any time, no matter we're holding rq_lock or not. For this reason
1094 * we allocate a separate ptlrpc_request and reply buffer for early reply
1097 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1098 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1099 * \a *req_ret to release it.
1100 * \retval -ev error number, and \a req_ret will not be set.
1102 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1103 struct ptlrpc_request **req_ret)
1105 struct ptlrpc_request *early_req;
1107 int early_bufsz, early_size;
1111 OBD_ALLOC_PTR(early_req);
1112 if (early_req == NULL)
1115 early_size = req->rq_nob_received;
1116 early_bufsz = size_roundup_power2(early_size);
1117 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1118 if (early_buf == NULL)
1119 GOTO(err_req, rc = -ENOMEM);
1121 /* sanity checkings and copy data out, do it inside spinlock */
1122 cfs_spin_lock(&req->rq_lock);
1124 if (req->rq_replied) {
1125 cfs_spin_unlock(&req->rq_lock);
1126 GOTO(err_buf, rc = -EALREADY);
1129 LASSERT(req->rq_repbuf);
1130 LASSERT(req->rq_repdata == NULL);
1131 LASSERT(req->rq_repmsg == NULL);
1133 if (req->rq_reply_off != 0) {
1134 CERROR("early reply with offset %u\n", req->rq_reply_off);
1135 cfs_spin_unlock(&req->rq_lock);
1136 GOTO(err_buf, rc = -EPROTO);
1139 if (req->rq_nob_received != early_size) {
1140 /* even another early arrived the size should be the same */
1141 CERROR("data size has changed from %u to %u\n",
1142 early_size, req->rq_nob_received);
1143 cfs_spin_unlock(&req->rq_lock);
1144 GOTO(err_buf, rc = -EINVAL);
1147 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1148 CERROR("early reply length %d too small\n",
1149 req->rq_nob_received);
1150 cfs_spin_unlock(&req->rq_lock);
1151 GOTO(err_buf, rc = -EALREADY);
1154 memcpy(early_buf, req->rq_repbuf, early_size);
1155 cfs_spin_unlock(&req->rq_lock);
1157 cfs_spin_lock_init(&early_req->rq_lock);
1158 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1159 early_req->rq_flvr = req->rq_flvr;
1160 early_req->rq_repbuf = early_buf;
1161 early_req->rq_repbuf_len = early_bufsz;
1162 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1163 early_req->rq_repdata_len = early_size;
1164 early_req->rq_early = 1;
1165 early_req->rq_reqmsg = req->rq_reqmsg;
1167 rc = do_cli_unwrap_reply(early_req);
1169 DEBUG_REQ(D_ADAPTTO, early_req,
1170 "error %d unwrap early reply", rc);
1174 LASSERT(early_req->rq_repmsg);
1175 *req_ret = early_req;
1179 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1181 OBD_FREE_LARGE(early_buf, early_bufsz);
1183 OBD_FREE_PTR(early_req);
1188 * Used by ptlrpc client, to release a processed early reply \a early_req.
1190 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1192 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1194 LASSERT(early_req->rq_repbuf);
1195 LASSERT(early_req->rq_repdata);
1196 LASSERT(early_req->rq_repmsg);
1198 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1199 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1200 OBD_FREE_PTR(early_req);
1203 /**************************************************
1205 **************************************************/
1208 * "fixed" sec (e.g. null) use sec_id < 0
1210 static cfs_atomic_t sptlrpc_sec_id = CFS_ATOMIC_INIT(1);
1212 int sptlrpc_get_next_secid(void)
1214 return cfs_atomic_inc_return(&sptlrpc_sec_id);
1216 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1218 /**************************************************
1219 * client side high-level security APIs *
1220 **************************************************/
1222 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1223 int grace, int force)
1225 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1227 LASSERT(policy->sp_cops);
1228 LASSERT(policy->sp_cops->flush_ctx_cache);
1230 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1233 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1235 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1237 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1238 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1239 LASSERT(policy->sp_cops->destroy_sec);
1241 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1243 policy->sp_cops->destroy_sec(sec);
1244 sptlrpc_policy_put(policy);
1247 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1249 sec_cop_destroy_sec(sec);
1251 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1253 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1255 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1257 if (sec->ps_policy->sp_cops->kill_sec) {
1258 sec->ps_policy->sp_cops->kill_sec(sec);
1260 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1264 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1267 cfs_atomic_inc(&sec->ps_refcount);
1271 EXPORT_SYMBOL(sptlrpc_sec_get);
1273 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1276 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1278 if (cfs_atomic_dec_and_test(&sec->ps_refcount)) {
1279 sptlrpc_gc_del_sec(sec);
1280 sec_cop_destroy_sec(sec);
1284 EXPORT_SYMBOL(sptlrpc_sec_put);
1287 * policy module is responsible for taking refrence of import
1290 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1291 struct ptlrpc_svc_ctx *svc_ctx,
1292 struct sptlrpc_flavor *sf,
1293 enum lustre_sec_part sp)
1295 struct ptlrpc_sec_policy *policy;
1296 struct ptlrpc_sec *sec;
1301 LASSERT(imp->imp_dlm_fake == 1);
1303 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1304 imp->imp_obd->obd_type->typ_name,
1305 imp->imp_obd->obd_name,
1306 sptlrpc_flavor2name(sf, str, sizeof(str)));
1308 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1309 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1311 LASSERT(imp->imp_dlm_fake == 0);
1313 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1314 imp->imp_obd->obd_type->typ_name,
1315 imp->imp_obd->obd_name,
1316 sptlrpc_flavor2name(sf, str, sizeof(str)));
1318 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1320 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1325 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1327 cfs_atomic_inc(&sec->ps_refcount);
1331 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1332 sptlrpc_gc_add_sec(sec);
1334 sptlrpc_policy_put(policy);
1340 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1342 struct ptlrpc_sec *sec;
1344 cfs_spin_lock(&imp->imp_lock);
1345 sec = sptlrpc_sec_get(imp->imp_sec);
1346 cfs_spin_unlock(&imp->imp_lock);
1350 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1352 static void sptlrpc_import_sec_install(struct obd_import *imp,
1353 struct ptlrpc_sec *sec)
1355 struct ptlrpc_sec *old_sec;
1357 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1359 cfs_spin_lock(&imp->imp_lock);
1360 old_sec = imp->imp_sec;
1362 cfs_spin_unlock(&imp->imp_lock);
1365 sptlrpc_sec_kill(old_sec);
1367 /* balance the ref taken by this import */
1368 sptlrpc_sec_put(old_sec);
1373 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1375 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1379 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1384 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1385 struct ptlrpc_sec *sec,
1386 struct sptlrpc_flavor *sf)
1388 char str1[32], str2[32];
1390 if (sec->ps_flvr.sf_flags != sf->sf_flags)
1391 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1392 sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1393 str1, sizeof(str1)),
1394 sptlrpc_secflags2str(sf->sf_flags,
1395 str2, sizeof(str2)));
1397 cfs_spin_lock(&sec->ps_lock);
1398 flavor_copy(&sec->ps_flvr, sf);
1399 cfs_spin_unlock(&sec->ps_lock);
1403 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1404 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1406 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1407 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1409 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1410 struct ptlrpc_svc_ctx *svc_ctx,
1411 struct sptlrpc_flavor *flvr)
1413 struct ptlrpc_connection *conn;
1414 struct sptlrpc_flavor sf;
1415 struct ptlrpc_sec *sec, *newsec;
1416 enum lustre_sec_part sp;
1426 conn = imp->imp_connection;
1428 if (svc_ctx == NULL) {
1429 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1431 * normal import, determine flavor from rule set, except
1432 * for mgc the flavor is predetermined.
1434 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1435 sf = cliobd->cl_flvr_mgc;
1437 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1439 &cliobd->cl_target_uuid,
1442 sp = imp->imp_obd->u.cli.cl_sp_me;
1444 /* reverse import, determine flavor from incoming reqeust */
1447 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1448 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1449 PTLRPC_SEC_FL_ROOTONLY;
1451 sp = sptlrpc_target_sec_part(imp->imp_obd);
1454 sec = sptlrpc_import_sec_ref(imp);
1458 if (flavor_equal(&sf, &sec->ps_flvr))
1461 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1462 imp->imp_obd->obd_name,
1463 obd_uuid2str(&conn->c_remote_uuid),
1464 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1465 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1467 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1468 SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1469 SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1470 SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1471 sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1474 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1475 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1476 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1477 imp->imp_obd->obd_name,
1478 obd_uuid2str(&conn->c_remote_uuid),
1479 LNET_NIDNET(conn->c_self),
1480 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1483 cfs_mutex_lock(&imp->imp_sec_mutex);
1485 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1487 sptlrpc_import_sec_install(imp, newsec);
1489 CERROR("import %s->%s: failed to create new sec\n",
1490 imp->imp_obd->obd_name,
1491 obd_uuid2str(&conn->c_remote_uuid));
1495 cfs_mutex_unlock(&imp->imp_sec_mutex);
1497 sptlrpc_sec_put(sec);
1501 void sptlrpc_import_sec_put(struct obd_import *imp)
1504 sptlrpc_sec_kill(imp->imp_sec);
1506 sptlrpc_sec_put(imp->imp_sec);
1507 imp->imp_sec = NULL;
1511 static void import_flush_ctx_common(struct obd_import *imp,
1512 uid_t uid, int grace, int force)
1514 struct ptlrpc_sec *sec;
1519 sec = sptlrpc_import_sec_ref(imp);
1523 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1524 sptlrpc_sec_put(sec);
1527 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1529 /* it's important to use grace mode, see explain in
1530 * sptlrpc_req_refresh_ctx() */
1531 import_flush_ctx_common(imp, 0, 1, 1);
1534 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1536 import_flush_ctx_common(imp, cfs_curproc_uid(), 1, 1);
1538 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1540 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1542 import_flush_ctx_common(imp, -1, 1, 1);
1544 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1547 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1548 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1550 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1552 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1553 struct ptlrpc_sec_policy *policy;
1557 LASSERT(ctx->cc_sec);
1558 LASSERT(ctx->cc_sec->ps_policy);
1559 LASSERT(req->rq_reqmsg == NULL);
1560 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1562 policy = ctx->cc_sec->ps_policy;
1563 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1565 LASSERT(req->rq_reqmsg);
1566 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1568 /* zeroing preallocated buffer */
1570 memset(req->rq_reqmsg, 0, msgsize);
1577 * Used by ptlrpc client to free request buffer of \a req. After this
1578 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1580 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1582 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1583 struct ptlrpc_sec_policy *policy;
1586 LASSERT(ctx->cc_sec);
1587 LASSERT(ctx->cc_sec->ps_policy);
1588 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1590 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1593 policy = ctx->cc_sec->ps_policy;
1594 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1595 req->rq_reqmsg = NULL;
1599 * NOTE caller must guarantee the buffer size is enough for the enlargement
1601 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1602 int segment, int newsize)
1605 int oldsize, oldmsg_size, movesize;
1607 LASSERT(segment < msg->lm_bufcount);
1608 LASSERT(msg->lm_buflens[segment] <= newsize);
1610 if (msg->lm_buflens[segment] == newsize)
1613 /* nothing to do if we are enlarging the last segment */
1614 if (segment == msg->lm_bufcount - 1) {
1615 msg->lm_buflens[segment] = newsize;
1619 oldsize = msg->lm_buflens[segment];
1621 src = lustre_msg_buf(msg, segment + 1, 0);
1622 msg->lm_buflens[segment] = newsize;
1623 dst = lustre_msg_buf(msg, segment + 1, 0);
1624 msg->lm_buflens[segment] = oldsize;
1626 /* move from segment + 1 to end segment */
1627 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1628 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1629 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1630 LASSERT(movesize >= 0);
1633 memmove(dst, src, movesize);
1635 /* note we don't clear the ares where old data live, not secret */
1637 /* finally set new segment size */
1638 msg->lm_buflens[segment] = newsize;
1640 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1643 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1644 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1645 * preserved after the enlargement. this must be called after original request
1646 * buffer being allocated.
1648 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1649 * so caller should refresh its local pointers if needed.
1651 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1652 int segment, int newsize)
1654 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1655 struct ptlrpc_sec_cops *cops;
1656 struct lustre_msg *msg = req->rq_reqmsg;
1660 LASSERT(msg->lm_bufcount > segment);
1661 LASSERT(msg->lm_buflens[segment] <= newsize);
1663 if (msg->lm_buflens[segment] == newsize)
1666 cops = ctx->cc_sec->ps_policy->sp_cops;
1667 LASSERT(cops->enlarge_reqbuf);
1668 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1670 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1673 * Used by ptlrpc client to allocate reply buffer of \a req.
1675 * \note After this, req->rq_repmsg is still not accessible.
1677 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1679 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1680 struct ptlrpc_sec_policy *policy;
1684 LASSERT(ctx->cc_sec);
1685 LASSERT(ctx->cc_sec->ps_policy);
1690 policy = ctx->cc_sec->ps_policy;
1691 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1695 * Used by ptlrpc client to free reply buffer of \a req. After this
1696 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1698 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1700 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1701 struct ptlrpc_sec_policy *policy;
1705 LASSERT(ctx->cc_sec);
1706 LASSERT(ctx->cc_sec->ps_policy);
1707 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1709 if (req->rq_repbuf == NULL)
1711 LASSERT(req->rq_repbuf_len);
1713 policy = ctx->cc_sec->ps_policy;
1714 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1715 req->rq_repmsg = NULL;
1719 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1720 struct ptlrpc_cli_ctx *ctx)
1722 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1724 if (!policy->sp_cops->install_rctx)
1726 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1729 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1730 struct ptlrpc_svc_ctx *ctx)
1732 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1734 if (!policy->sp_sops->install_rctx)
1736 return policy->sp_sops->install_rctx(imp, ctx);
1739 /****************************************
1740 * server side security *
1741 ****************************************/
1743 static int flavor_allowed(struct sptlrpc_flavor *exp,
1744 struct ptlrpc_request *req)
1746 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1748 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1751 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1752 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1753 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1754 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1760 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1763 * Given an export \a exp, check whether the flavor of incoming \a req
1764 * is allowed by the export \a exp. Main logic is about taking care of
1765 * changing configurations. Return 0 means success.
1767 int sptlrpc_target_export_check(struct obd_export *exp,
1768 struct ptlrpc_request *req)
1770 struct sptlrpc_flavor flavor;
1775 /* client side export has no imp_reverse, skip
1776 * FIXME maybe we should check flavor this as well??? */
1777 if (exp->exp_imp_reverse == NULL)
1780 /* don't care about ctx fini rpc */
1781 if (req->rq_ctx_fini)
1784 cfs_spin_lock(&exp->exp_lock);
1786 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1787 * the first req with the new flavor, then treat it as current flavor,
1788 * adapt reverse sec according to it.
1789 * note the first rpc with new flavor might not be with root ctx, in
1790 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1791 if (unlikely(exp->exp_flvr_changed) &&
1792 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1793 /* make the new flavor as "current", and old ones as
1794 * about-to-expire */
1795 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1796 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1797 flavor = exp->exp_flvr_old[1];
1798 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1799 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1800 exp->exp_flvr_old[0] = exp->exp_flvr;
1801 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1802 EXP_FLVR_UPDATE_EXPIRE;
1803 exp->exp_flvr = flavor;
1805 /* flavor change finished */
1806 exp->exp_flvr_changed = 0;
1807 LASSERT(exp->exp_flvr_adapt == 1);
1809 /* if it's gss, we only interested in root ctx init */
1810 if (req->rq_auth_gss &&
1811 !(req->rq_ctx_init &&
1812 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1813 req->rq_auth_usr_ost))) {
1814 cfs_spin_unlock(&exp->exp_lock);
1815 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1816 req->rq_auth_gss, req->rq_ctx_init,
1817 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1818 req->rq_auth_usr_ost);
1822 exp->exp_flvr_adapt = 0;
1823 cfs_spin_unlock(&exp->exp_lock);
1825 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1826 req->rq_svc_ctx, &flavor);
1829 /* if it equals to the current flavor, we accept it, but need to
1830 * dealing with reverse sec/ctx */
1831 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1832 /* most cases should return here, we only interested in
1833 * gss root ctx init */
1834 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1835 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1836 !req->rq_auth_usr_ost)) {
1837 cfs_spin_unlock(&exp->exp_lock);
1841 /* if flavor just changed, we should not proceed, just leave
1842 * it and current flavor will be discovered and replaced
1843 * shortly, and let _this_ rpc pass through */
1844 if (exp->exp_flvr_changed) {
1845 LASSERT(exp->exp_flvr_adapt);
1846 cfs_spin_unlock(&exp->exp_lock);
1850 if (exp->exp_flvr_adapt) {
1851 exp->exp_flvr_adapt = 0;
1852 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1853 exp, exp->exp_flvr.sf_rpc,
1854 exp->exp_flvr_old[0].sf_rpc,
1855 exp->exp_flvr_old[1].sf_rpc);
1856 flavor = exp->exp_flvr;
1857 cfs_spin_unlock(&exp->exp_lock);
1859 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1863 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1864 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1865 exp->exp_flvr_old[0].sf_rpc,
1866 exp->exp_flvr_old[1].sf_rpc);
1867 cfs_spin_unlock(&exp->exp_lock);
1869 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1874 if (exp->exp_flvr_expire[0]) {
1875 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1876 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1877 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1878 "middle one ("CFS_DURATION_T")\n", exp,
1879 exp->exp_flvr.sf_rpc,
1880 exp->exp_flvr_old[0].sf_rpc,
1881 exp->exp_flvr_old[1].sf_rpc,
1882 exp->exp_flvr_expire[0] -
1883 cfs_time_current_sec());
1884 cfs_spin_unlock(&exp->exp_lock);
1888 CDEBUG(D_SEC, "mark middle expired\n");
1889 exp->exp_flvr_expire[0] = 0;
1891 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1892 exp->exp_flvr.sf_rpc,
1893 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1894 req->rq_flvr.sf_rpc);
1897 /* now it doesn't match the current flavor, the only chance we can
1898 * accept it is match the old flavors which is not expired. */
1899 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1900 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1901 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1902 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1903 "oldest one ("CFS_DURATION_T")\n", exp,
1904 exp->exp_flvr.sf_rpc,
1905 exp->exp_flvr_old[0].sf_rpc,
1906 exp->exp_flvr_old[1].sf_rpc,
1907 exp->exp_flvr_expire[1] -
1908 cfs_time_current_sec());
1909 cfs_spin_unlock(&exp->exp_lock);
1913 CDEBUG(D_SEC, "mark oldest expired\n");
1914 exp->exp_flvr_expire[1] = 0;
1916 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1917 exp, exp->exp_flvr.sf_rpc,
1918 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1919 req->rq_flvr.sf_rpc);
1921 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1922 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1923 exp->exp_flvr_old[1].sf_rpc);
1926 cfs_spin_unlock(&exp->exp_lock);
1928 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1929 "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1930 exp, exp->exp_obd->obd_name,
1931 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1932 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1933 req->rq_flvr.sf_rpc,
1934 exp->exp_flvr.sf_rpc,
1935 exp->exp_flvr_old[0].sf_rpc,
1936 exp->exp_flvr_expire[0] ?
1937 (unsigned long) (exp->exp_flvr_expire[0] -
1938 cfs_time_current_sec()) : 0,
1939 exp->exp_flvr_old[1].sf_rpc,
1940 exp->exp_flvr_expire[1] ?
1941 (unsigned long) (exp->exp_flvr_expire[1] -
1942 cfs_time_current_sec()) : 0);
1945 EXPORT_SYMBOL(sptlrpc_target_export_check);
1947 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1948 struct sptlrpc_rule_set *rset)
1950 struct obd_export *exp;
1951 struct sptlrpc_flavor new_flvr;
1955 cfs_spin_lock(&obd->obd_dev_lock);
1957 cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1958 if (exp->exp_connection == NULL)
1961 /* note if this export had just been updated flavor
1962 * (exp_flvr_changed == 1), this will override the
1964 cfs_spin_lock(&exp->exp_lock);
1965 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1966 exp->exp_connection->c_peer.nid,
1968 if (exp->exp_flvr_changed ||
1969 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1970 exp->exp_flvr_old[1] = new_flvr;
1971 exp->exp_flvr_expire[1] = 0;
1972 exp->exp_flvr_changed = 1;
1973 exp->exp_flvr_adapt = 1;
1975 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1976 exp, sptlrpc_part2name(exp->exp_sp_peer),
1977 exp->exp_flvr.sf_rpc,
1978 exp->exp_flvr_old[1].sf_rpc);
1980 cfs_spin_unlock(&exp->exp_lock);
1983 cfs_spin_unlock(&obd->obd_dev_lock);
1985 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1987 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1989 /* peer's claim is unreliable unless gss is being used */
1990 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1993 switch (req->rq_sp_from) {
1995 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1996 DEBUG_REQ(D_ERROR, req, "faked source CLI");
1997 svc_rc = SECSVC_DROP;
2001 if (!req->rq_auth_usr_mdt) {
2002 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2003 svc_rc = SECSVC_DROP;
2007 if (!req->rq_auth_usr_ost) {
2008 DEBUG_REQ(D_ERROR, req, "faked source OST");
2009 svc_rc = SECSVC_DROP;
2014 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2015 !req->rq_auth_usr_ost) {
2016 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2017 svc_rc = SECSVC_DROP;
2022 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2023 svc_rc = SECSVC_DROP;
2030 * Used by ptlrpc server, to perform transformation upon request message of
2031 * incoming \a req. This must be the first thing to do with a incoming
2032 * request in ptlrpc layer.
2034 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2035 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2036 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2037 * reply message has been prepared.
2038 * \retval SECSVC_DROP failed, this request should be dropped.
2040 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2042 struct ptlrpc_sec_policy *policy;
2043 struct lustre_msg *msg = req->rq_reqbuf;
2048 LASSERT(req->rq_reqmsg == NULL);
2049 LASSERT(req->rq_repmsg == NULL);
2050 LASSERT(req->rq_svc_ctx == NULL);
2052 req->rq_req_swab_mask = 0;
2054 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2057 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2061 CERROR("error unpacking request from %s x"LPU64"\n",
2062 libcfs_id2str(req->rq_peer), req->rq_xid);
2063 RETURN(SECSVC_DROP);
2066 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2067 req->rq_sp_from = LUSTRE_SP_ANY;
2068 req->rq_auth_uid = INVALID_UID;
2069 req->rq_auth_mapped_uid = INVALID_UID;
2071 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2073 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2074 RETURN(SECSVC_DROP);
2077 LASSERT(policy->sp_sops->accept);
2078 rc = policy->sp_sops->accept(req);
2079 sptlrpc_policy_put(policy);
2080 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2081 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2084 * if it's not null flavor (which means embedded packing msg),
2085 * reset the swab mask for the comming inner msg unpacking.
2087 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2088 req->rq_req_swab_mask = 0;
2090 /* sanity check for the request source */
2091 rc = sptlrpc_svc_check_from(req, rc);
2096 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2097 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2098 * a buffer of \a msglen size.
2100 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2102 struct ptlrpc_sec_policy *policy;
2103 struct ptlrpc_reply_state *rs;
2107 LASSERT(req->rq_svc_ctx);
2108 LASSERT(req->rq_svc_ctx->sc_policy);
2110 policy = req->rq_svc_ctx->sc_policy;
2111 LASSERT(policy->sp_sops->alloc_rs);
2113 rc = policy->sp_sops->alloc_rs(req, msglen);
2114 if (unlikely(rc == -ENOMEM)) {
2115 /* failed alloc, try emergency pool */
2116 rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service);
2120 req->rq_reply_state = rs;
2121 rc = policy->sp_sops->alloc_rs(req, msglen);
2123 lustre_put_emerg_rs(rs);
2124 req->rq_reply_state = NULL;
2129 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2135 * Used by ptlrpc server, to perform transformation upon reply message.
2137 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2138 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2140 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2142 struct ptlrpc_sec_policy *policy;
2146 LASSERT(req->rq_svc_ctx);
2147 LASSERT(req->rq_svc_ctx->sc_policy);
2149 policy = req->rq_svc_ctx->sc_policy;
2150 LASSERT(policy->sp_sops->authorize);
2152 rc = policy->sp_sops->authorize(req);
2153 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2159 * Used by ptlrpc server, to free reply_state.
2161 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2163 struct ptlrpc_sec_policy *policy;
2164 unsigned int prealloc;
2167 LASSERT(rs->rs_svc_ctx);
2168 LASSERT(rs->rs_svc_ctx->sc_policy);
2170 policy = rs->rs_svc_ctx->sc_policy;
2171 LASSERT(policy->sp_sops->free_rs);
2173 prealloc = rs->rs_prealloc;
2174 policy->sp_sops->free_rs(rs);
2177 lustre_put_emerg_rs(rs);
2181 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2183 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2186 cfs_atomic_inc(&ctx->sc_refcount);
2189 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2191 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2196 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2197 if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) {
2198 if (ctx->sc_policy->sp_sops->free_ctx)
2199 ctx->sc_policy->sp_sops->free_ctx(ctx);
2201 req->rq_svc_ctx = NULL;
2204 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2206 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2211 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2212 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2213 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2215 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2217 /****************************************
2219 ****************************************/
2222 * Perform transformation upon bulk data pointed by \a desc. This is called
2223 * before transforming the request message.
2225 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2226 struct ptlrpc_bulk_desc *desc)
2228 struct ptlrpc_cli_ctx *ctx;
2230 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2232 if (!req->rq_pack_bulk)
2235 ctx = req->rq_cli_ctx;
2236 if (ctx->cc_ops->wrap_bulk)
2237 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2240 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2243 * This is called after unwrap the reply message.
2244 * return nob of actual plain text size received, or error code.
2246 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2247 struct ptlrpc_bulk_desc *desc,
2250 struct ptlrpc_cli_ctx *ctx;
2253 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2255 if (!req->rq_pack_bulk)
2256 return desc->bd_nob_transferred;
2258 ctx = req->rq_cli_ctx;
2259 if (ctx->cc_ops->unwrap_bulk) {
2260 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2264 return desc->bd_nob_transferred;
2266 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2269 * This is called after unwrap the reply message.
2270 * return 0 for success or error code.
2272 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2273 struct ptlrpc_bulk_desc *desc)
2275 struct ptlrpc_cli_ctx *ctx;
2278 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2280 if (!req->rq_pack_bulk)
2283 ctx = req->rq_cli_ctx;
2284 if (ctx->cc_ops->unwrap_bulk) {
2285 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2291 * if everything is going right, nob should equals to nob_transferred.
2292 * in case of privacy mode, nob_transferred needs to be adjusted.
2294 if (desc->bd_nob != desc->bd_nob_transferred) {
2295 CERROR("nob %d doesn't match transferred nob %d",
2296 desc->bd_nob, desc->bd_nob_transferred);
2302 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2304 #ifdef HAVE_SERVER_SUPPORT
2306 * Performe transformation upon outgoing bulk read.
2308 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2309 struct ptlrpc_bulk_desc *desc)
2311 struct ptlrpc_svc_ctx *ctx;
2313 LASSERT(req->rq_bulk_read);
2315 if (!req->rq_pack_bulk)
2318 ctx = req->rq_svc_ctx;
2319 if (ctx->sc_policy->sp_sops->wrap_bulk)
2320 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2324 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2327 * Performe transformation upon incoming bulk write.
2329 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2330 struct ptlrpc_bulk_desc *desc)
2332 struct ptlrpc_svc_ctx *ctx;
2335 LASSERT(req->rq_bulk_write);
2338 * if it's in privacy mode, transferred should >= expected; otherwise
2339 * transferred should == expected.
2341 if (desc->bd_nob_transferred < desc->bd_nob ||
2342 (desc->bd_nob_transferred > desc->bd_nob &&
2343 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2344 SPTLRPC_BULK_SVC_PRIV)) {
2345 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2346 desc->bd_nob_transferred, desc->bd_nob);
2350 if (!req->rq_pack_bulk)
2353 ctx = req->rq_svc_ctx;
2354 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2355 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2357 CERROR("error unwrap bulk: %d\n", rc);
2360 /* return 0 to allow reply be sent */
2363 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2366 * Prepare buffers for incoming bulk write.
2368 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2369 struct ptlrpc_bulk_desc *desc)
2371 struct ptlrpc_svc_ctx *ctx;
2373 LASSERT(req->rq_bulk_write);
2375 if (!req->rq_pack_bulk)
2378 ctx = req->rq_svc_ctx;
2379 if (ctx->sc_policy->sp_sops->prep_bulk)
2380 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2384 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2386 #endif /* HAVE_SERVER_SUPPORT */
2388 /****************************************
2389 * user descriptor helpers *
2390 ****************************************/
2392 int sptlrpc_current_user_desc_size(void)
2397 ngroups = current_ngroups;
2399 if (ngroups > LUSTRE_MAX_GROUPS)
2400 ngroups = LUSTRE_MAX_GROUPS;
2404 return sptlrpc_user_desc_size(ngroups);
2406 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2408 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2410 struct ptlrpc_user_desc *pud;
2412 pud = lustre_msg_buf(msg, offset, 0);
2414 pud->pud_uid = cfs_curproc_uid();
2415 pud->pud_gid = cfs_curproc_gid();
2416 pud->pud_fsuid = cfs_curproc_fsuid();
2417 pud->pud_fsgid = cfs_curproc_fsgid();
2418 pud->pud_cap = cfs_curproc_cap_pack();
2419 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2423 if (pud->pud_ngroups > current_ngroups)
2424 pud->pud_ngroups = current_ngroups;
2425 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2426 pud->pud_ngroups * sizeof(__u32));
2427 task_unlock(current);
2432 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2434 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2436 struct ptlrpc_user_desc *pud;
2439 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2444 __swab32s(&pud->pud_uid);
2445 __swab32s(&pud->pud_gid);
2446 __swab32s(&pud->pud_fsuid);
2447 __swab32s(&pud->pud_fsgid);
2448 __swab32s(&pud->pud_cap);
2449 __swab32s(&pud->pud_ngroups);
2452 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2453 CERROR("%u groups is too large\n", pud->pud_ngroups);
2457 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2458 msg->lm_buflens[offset]) {
2459 CERROR("%u groups are claimed but bufsize only %u\n",
2460 pud->pud_ngroups, msg->lm_buflens[offset]);
2465 for (i = 0; i < pud->pud_ngroups; i++)
2466 __swab32s(&pud->pud_groups[i]);
2471 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2473 /****************************************
2475 ****************************************/
2477 const char * sec2target_str(struct ptlrpc_sec *sec)
2479 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2481 if (sec_is_reverse(sec))
2483 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2485 EXPORT_SYMBOL(sec2target_str);
2488 * return true if the bulk data is protected
2490 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2492 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2493 case SPTLRPC_BULK_SVC_INTG:
2494 case SPTLRPC_BULK_SVC_PRIV:
2500 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2502 /****************************************
2503 * crypto API helper/alloc blkciper *
2504 ****************************************/
2506 /****************************************
2507 * initialize/finalize *
2508 ****************************************/
2510 int sptlrpc_init(void)
2514 cfs_rwlock_init(&policy_lock);
2516 rc = sptlrpc_gc_init();
2520 rc = sptlrpc_conf_init();
2524 rc = sptlrpc_enc_pool_init();
2528 rc = sptlrpc_null_init();
2532 rc = sptlrpc_plain_init();
2536 rc = sptlrpc_lproc_init();
2543 sptlrpc_plain_fini();
2545 sptlrpc_null_fini();
2547 sptlrpc_enc_pool_fini();
2549 sptlrpc_conf_fini();
2556 void sptlrpc_fini(void)
2558 sptlrpc_lproc_fini();
2559 sptlrpc_plain_fini();
2560 sptlrpc_null_fini();
2561 sptlrpc_enc_pool_fini();
2562 sptlrpc_conf_fini();