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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Eric Mei <ericm@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_SEC
43 #include <libcfs/libcfs.h>
44 #include <linux/crypto.h>
45 #include <linux/key.h>
48 #include <obd_class.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
51 #include <lustre_import.h>
52 #include <lustre_dlm.h>
53 #include <lustre_sec.h>
55 #include "ptlrpc_internal.h"
57 /***********************************************
59 ***********************************************/
61 static rwlock_t policy_lock;
62 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
66 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
68 __u16 number = policy->sp_policy;
70 LASSERT(policy->sp_name);
71 LASSERT(policy->sp_cops);
72 LASSERT(policy->sp_sops);
74 if (number >= SPTLRPC_POLICY_MAX)
77 write_lock(&policy_lock);
78 if (unlikely(policies[number])) {
79 write_unlock(&policy_lock);
82 policies[number] = policy;
83 write_unlock(&policy_lock);
85 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
88 EXPORT_SYMBOL(sptlrpc_register_policy);
90 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
92 __u16 number = policy->sp_policy;
94 LASSERT(number < SPTLRPC_POLICY_MAX);
96 write_lock(&policy_lock);
97 if (unlikely(policies[number] == NULL)) {
98 write_unlock(&policy_lock);
99 CERROR("%s: already unregistered\n", policy->sp_name);
103 LASSERT(policies[number] == policy);
104 policies[number] = NULL;
105 write_unlock(&policy_lock);
107 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
110 EXPORT_SYMBOL(sptlrpc_unregister_policy);
113 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
115 static DEFINE_MUTEX(load_mutex);
116 static atomic_t loaded = ATOMIC_INIT(0);
117 struct ptlrpc_sec_policy *policy;
118 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
121 if (number >= SPTLRPC_POLICY_MAX)
125 read_lock(&policy_lock);
126 policy = policies[number];
127 if (policy && !try_module_get(policy->sp_owner))
130 flag = atomic_read(&loaded);
131 read_unlock(&policy_lock);
133 if (policy != NULL || flag != 0 ||
134 number != SPTLRPC_POLICY_GSS)
137 /* try to load gss module, once */
138 mutex_lock(&load_mutex);
139 if (atomic_read(&loaded) == 0) {
140 if (request_module("ptlrpc_gss") == 0)
142 "module ptlrpc_gss loaded on demand\n");
144 CERROR("Unable to load module ptlrpc_gss\n");
146 atomic_set(&loaded, 1);
148 mutex_unlock(&load_mutex);
154 __u32 sptlrpc_name2flavor_base(const char *name)
156 if (!strcmp(name, "null"))
157 return SPTLRPC_FLVR_NULL;
158 if (!strcmp(name, "plain"))
159 return SPTLRPC_FLVR_PLAIN;
160 if (!strcmp(name, "gssnull"))
161 return SPTLRPC_FLVR_GSSNULL;
162 if (!strcmp(name, "krb5n"))
163 return SPTLRPC_FLVR_KRB5N;
164 if (!strcmp(name, "krb5a"))
165 return SPTLRPC_FLVR_KRB5A;
166 if (!strcmp(name, "krb5i"))
167 return SPTLRPC_FLVR_KRB5I;
168 if (!strcmp(name, "krb5p"))
169 return SPTLRPC_FLVR_KRB5P;
170 if (!strcmp(name, "ski"))
171 return SPTLRPC_FLVR_SKI;
172 if (!strcmp(name, "skpi"))
173 return SPTLRPC_FLVR_SKPI;
175 return SPTLRPC_FLVR_INVALID;
177 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
179 const char *sptlrpc_flavor2name_base(__u32 flvr)
181 __u32 base = SPTLRPC_FLVR_BASE(flvr);
183 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
185 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
187 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
189 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
191 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
199 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
202 CERROR("invalid wire flavor 0x%x\n", flvr);
205 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
207 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
208 char *buf, int bufsize)
210 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
211 snprintf(buf, bufsize, "hash:%s",
212 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
214 snprintf(buf, bufsize, "%s",
215 sptlrpc_flavor2name_base(sf->sf_rpc));
217 buf[bufsize - 1] = '\0';
220 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
222 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
224 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
227 * currently we don't support customized bulk specification for
228 * flavors other than plain
230 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
234 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
235 strncat(buf, bspec, bufsize);
238 buf[bufsize - 1] = '\0';
241 EXPORT_SYMBOL(sptlrpc_flavor2name);
243 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
247 if (flags & PTLRPC_SEC_FL_REVERSE)
248 strlcat(buf, "reverse,", bufsize);
249 if (flags & PTLRPC_SEC_FL_ROOTONLY)
250 strlcat(buf, "rootonly,", bufsize);
251 if (flags & PTLRPC_SEC_FL_UDESC)
252 strlcat(buf, "udesc,", bufsize);
253 if (flags & PTLRPC_SEC_FL_BULK)
254 strlcat(buf, "bulk,", bufsize);
256 strlcat(buf, "-,", bufsize);
260 EXPORT_SYMBOL(sptlrpc_secflags2str);
262 /**************************************************
263 * client context APIs *
264 **************************************************/
267 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
269 struct vfs_cred vcred;
270 int create = 1, remove_dead = 1;
273 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
275 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
276 PTLRPC_SEC_FL_ROOTONLY)) {
279 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
284 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
285 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
288 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
292 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
294 atomic_inc(&ctx->cc_refcount);
297 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
299 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
301 struct ptlrpc_sec *sec = ctx->cc_sec;
304 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
306 if (!atomic_dec_and_test(&ctx->cc_refcount))
309 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
311 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
314 * Expire the client context immediately.
316 * \pre Caller must hold at least 1 reference on the \a ctx.
318 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
320 LASSERT(ctx->cc_ops->die);
321 ctx->cc_ops->die(ctx, 0);
323 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
326 * To wake up the threads who are waiting for this client context. Called
327 * after some status change happened on \a ctx.
329 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
331 struct ptlrpc_request *req, *next;
333 spin_lock(&ctx->cc_lock);
334 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
336 list_del_init(&req->rq_ctx_chain);
337 ptlrpc_client_wake_req(req);
339 spin_unlock(&ctx->cc_lock);
341 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
343 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
345 LASSERT(ctx->cc_ops);
347 if (ctx->cc_ops->display == NULL)
350 return ctx->cc_ops->display(ctx, buf, bufsize);
353 static int import_sec_check_expire(struct obd_import *imp)
357 spin_lock(&imp->imp_lock);
358 if (imp->imp_sec_expire &&
359 imp->imp_sec_expire < cfs_time_current_sec()) {
361 imp->imp_sec_expire = 0;
363 spin_unlock(&imp->imp_lock);
368 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
369 return sptlrpc_import_sec_adapt(imp, NULL, 0);
372 static int import_sec_validate_get(struct obd_import *imp,
373 struct ptlrpc_sec **sec)
377 if (unlikely(imp->imp_sec_expire)) {
378 rc = import_sec_check_expire(imp);
383 *sec = sptlrpc_import_sec_ref(imp);
385 CERROR("import %p (%s) with no sec\n",
386 imp, ptlrpc_import_state_name(imp->imp_state));
390 if (unlikely((*sec)->ps_dying)) {
391 CERROR("attempt to use dying sec %p\n", sec);
392 sptlrpc_sec_put(*sec);
400 * Given a \a req, find or allocate a appropriate context for it.
401 * \pre req->rq_cli_ctx == NULL.
403 * \retval 0 succeed, and req->rq_cli_ctx is set.
404 * \retval -ev error number, and req->rq_cli_ctx == NULL.
406 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
408 struct obd_import *imp = req->rq_import;
409 struct ptlrpc_sec *sec;
413 LASSERT(!req->rq_cli_ctx);
416 rc = import_sec_validate_get(imp, &sec);
420 req->rq_cli_ctx = get_my_ctx(sec);
422 sptlrpc_sec_put(sec);
424 if (!req->rq_cli_ctx) {
425 CERROR("req %p: fail to get context\n", req);
433 * Drop the context for \a req.
434 * \pre req->rq_cli_ctx != NULL.
435 * \post req->rq_cli_ctx == NULL.
437 * If \a sync == 0, this function should return quickly without sleep;
438 * otherwise it might trigger and wait for the whole process of sending
439 * an context-destroying rpc to server.
441 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
446 LASSERT(req->rq_cli_ctx);
448 /* request might be asked to release earlier while still
449 * in the context waiting list.
451 if (!list_empty(&req->rq_ctx_chain)) {
452 spin_lock(&req->rq_cli_ctx->cc_lock);
453 list_del_init(&req->rq_ctx_chain);
454 spin_unlock(&req->rq_cli_ctx->cc_lock);
457 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
458 req->rq_cli_ctx = NULL;
463 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
464 struct ptlrpc_cli_ctx *oldctx,
465 struct ptlrpc_cli_ctx *newctx)
467 struct sptlrpc_flavor old_flvr;
468 char *reqmsg = NULL; /* to workaround old gcc */
472 LASSERT(req->rq_reqmsg);
473 LASSERT(req->rq_reqlen);
474 LASSERT(req->rq_replen);
476 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
477 "switch sec %p(%s) -> %p(%s)\n", req,
478 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
479 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
480 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
481 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
484 old_flvr = req->rq_flvr;
486 /* save request message */
487 reqmsg_size = req->rq_reqlen;
488 if (reqmsg_size != 0) {
489 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
492 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
495 /* release old req/rep buf */
496 req->rq_cli_ctx = oldctx;
497 sptlrpc_cli_free_reqbuf(req);
498 sptlrpc_cli_free_repbuf(req);
499 req->rq_cli_ctx = newctx;
501 /* recalculate the flavor */
502 sptlrpc_req_set_flavor(req, 0);
504 /* alloc new request buffer
505 * we don't need to alloc reply buffer here, leave it to the
506 * rest procedure of ptlrpc */
507 if (reqmsg_size != 0) {
508 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
510 LASSERT(req->rq_reqmsg);
511 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
513 CWARN("failed to alloc reqbuf: %d\n", rc);
514 req->rq_flvr = old_flvr;
517 OBD_FREE_LARGE(reqmsg, reqmsg_size);
523 * If current context of \a req is dead somehow, e.g. we just switched flavor
524 * thus marked original contexts dead, we'll find a new context for it. if
525 * no switch is needed, \a req will end up with the same context.
527 * \note a request must have a context, to keep other parts of code happy.
528 * In any case of failure during the switching, we must restore the old one.
530 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
532 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
533 struct ptlrpc_cli_ctx *newctx;
539 sptlrpc_cli_ctx_get(oldctx);
540 sptlrpc_req_put_ctx(req, 0);
542 rc = sptlrpc_req_get_ctx(req);
544 LASSERT(!req->rq_cli_ctx);
546 /* restore old ctx */
547 req->rq_cli_ctx = oldctx;
551 newctx = req->rq_cli_ctx;
554 if (unlikely(newctx == oldctx &&
555 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
557 * still get the old dead ctx, usually means system too busy
560 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
561 newctx, newctx->cc_flags);
563 schedule_timeout_and_set_state(TASK_INTERRUPTIBLE,
567 * it's possible newctx == oldctx if we're switching
568 * subflavor with the same sec.
570 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
572 /* restore old ctx */
573 sptlrpc_req_put_ctx(req, 0);
574 req->rq_cli_ctx = oldctx;
578 LASSERT(req->rq_cli_ctx == newctx);
581 sptlrpc_cli_ctx_put(oldctx, 1);
584 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
587 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
589 if (cli_ctx_is_refreshed(ctx))
595 int ctx_refresh_timeout(void *data)
597 struct ptlrpc_request *req = data;
600 /* conn_cnt is needed in expire_one_request */
601 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
603 rc = ptlrpc_expire_one_request(req, 1);
604 /* if we started recovery, we should mark this ctx dead; otherwise
605 * in case of lgssd died nobody would retire this ctx, following
606 * connecting will still find the same ctx thus cause deadlock.
607 * there's an assumption that expire time of the request should be
608 * later than the context refresh expire time.
611 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
616 void ctx_refresh_interrupt(void *data)
618 struct ptlrpc_request *req = data;
620 spin_lock(&req->rq_lock);
622 spin_unlock(&req->rq_lock);
626 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
628 spin_lock(&ctx->cc_lock);
629 if (!list_empty(&req->rq_ctx_chain))
630 list_del_init(&req->rq_ctx_chain);
631 spin_unlock(&ctx->cc_lock);
635 * To refresh the context of \req, if it's not up-to-date.
638 * - = 0: wait until success or fatal error occur
639 * - > 0: timeout value (in seconds)
641 * The status of the context could be subject to be changed by other threads
642 * at any time. We allow this race, but once we return with 0, the caller will
643 * suppose it's uptodated and keep using it until the owning rpc is done.
645 * \retval 0 only if the context is uptodated.
646 * \retval -ev error number.
648 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
650 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
651 struct ptlrpc_sec *sec;
652 struct l_wait_info lwi;
658 if (req->rq_ctx_init || req->rq_ctx_fini)
662 * during the process a request's context might change type even
663 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
667 rc = import_sec_validate_get(req->rq_import, &sec);
671 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
672 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
673 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
674 req_off_ctx_list(req, ctx);
675 sptlrpc_req_replace_dead_ctx(req);
676 ctx = req->rq_cli_ctx;
678 sptlrpc_sec_put(sec);
680 if (cli_ctx_is_eternal(ctx))
683 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
684 LASSERT(ctx->cc_ops->refresh);
685 ctx->cc_ops->refresh(ctx);
687 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
689 LASSERT(ctx->cc_ops->validate);
690 if (ctx->cc_ops->validate(ctx) == 0) {
691 req_off_ctx_list(req, ctx);
695 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
696 spin_lock(&req->rq_lock);
698 spin_unlock(&req->rq_lock);
699 req_off_ctx_list(req, ctx);
704 * There's a subtle issue for resending RPCs, suppose following
706 * 1. the request was sent to server.
707 * 2. recovery was kicked start, after finished the request was
709 * 3. resend the request.
710 * 4. old reply from server received, we accept and verify the reply.
711 * this has to be success, otherwise the error will be aware
713 * 5. new reply from server received, dropped by LNet.
715 * Note the xid of old & new request is the same. We can't simply
716 * change xid for the resent request because the server replies on
717 * it for reply reconstruction.
719 * Commonly the original context should be uptodate because we
720 * have a expiry nice time; server will keep its context because
721 * we at least hold a ref of old context which prevent context
722 * destroying RPC being sent. So server still can accept the request
723 * and finish the RPC. But if that's not the case:
724 * 1. If server side context has been trimmed, a NO_CONTEXT will
725 * be returned, gss_cli_ctx_verify/unseal will switch to new
727 * 2. Current context never be refreshed, then we are fine: we
728 * never really send request with old context before.
730 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
731 unlikely(req->rq_reqmsg) &&
732 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
733 req_off_ctx_list(req, ctx);
737 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
738 req_off_ctx_list(req, ctx);
740 * don't switch ctx if import was deactivated
742 if (req->rq_import->imp_deactive) {
743 spin_lock(&req->rq_lock);
745 spin_unlock(&req->rq_lock);
749 rc = sptlrpc_req_replace_dead_ctx(req);
751 LASSERT(ctx == req->rq_cli_ctx);
752 CERROR("req %p: failed to replace dead ctx %p: %d\n",
754 spin_lock(&req->rq_lock);
756 spin_unlock(&req->rq_lock);
760 ctx = req->rq_cli_ctx;
765 * Now we're sure this context is during upcall, add myself into
768 spin_lock(&ctx->cc_lock);
769 if (list_empty(&req->rq_ctx_chain))
770 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
771 spin_unlock(&ctx->cc_lock);
774 RETURN(-EWOULDBLOCK);
776 /* Clear any flags that may be present from previous sends */
777 LASSERT(req->rq_receiving_reply == 0);
778 spin_lock(&req->rq_lock);
780 req->rq_timedout = 0;
783 spin_unlock(&req->rq_lock);
785 lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout,
786 ctx_refresh_interrupt, req);
787 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
790 * following cases could lead us here:
791 * - successfully refreshed;
793 * - timedout, and we don't want recover from the failure;
794 * - timedout, and waked up upon recovery finished;
795 * - someone else mark this ctx dead by force;
796 * - someone invalidate the req and call ptlrpc_client_wake_req(),
797 * e.g. ptlrpc_abort_inflight();
799 if (!cli_ctx_is_refreshed(ctx)) {
800 /* timed out or interruptted */
801 req_off_ctx_list(req, ctx);
811 * Initialize flavor settings for \a req, according to \a opcode.
813 * \note this could be called in two situations:
814 * - new request from ptlrpc_pre_req(), with proper @opcode
815 * - old request which changed ctx in the middle, with @opcode == 0
817 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
819 struct ptlrpc_sec *sec;
821 LASSERT(req->rq_import);
822 LASSERT(req->rq_cli_ctx);
823 LASSERT(req->rq_cli_ctx->cc_sec);
824 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
826 /* special security flags accoding to opcode */
830 case MGS_CONFIG_READ:
832 req->rq_bulk_read = 1;
836 req->rq_bulk_write = 1;
839 req->rq_ctx_init = 1;
842 req->rq_ctx_fini = 1;
845 /* init/fini rpc won't be resend, so can't be here */
846 LASSERT(req->rq_ctx_init == 0);
847 LASSERT(req->rq_ctx_fini == 0);
849 /* cleanup flags, which should be recalculated */
850 req->rq_pack_udesc = 0;
851 req->rq_pack_bulk = 0;
855 sec = req->rq_cli_ctx->cc_sec;
857 spin_lock(&sec->ps_lock);
858 req->rq_flvr = sec->ps_flvr;
859 spin_unlock(&sec->ps_lock);
861 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
863 if (unlikely(req->rq_ctx_init))
864 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
865 else if (unlikely(req->rq_ctx_fini))
866 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
868 /* user descriptor flag, null security can't do it anyway */
869 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
870 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
871 req->rq_pack_udesc = 1;
873 /* bulk security flag */
874 if ((req->rq_bulk_read || req->rq_bulk_write) &&
875 sptlrpc_flavor_has_bulk(&req->rq_flvr))
876 req->rq_pack_bulk = 1;
879 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
881 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
884 LASSERT(req->rq_clrbuf);
885 if (req->rq_pool || !req->rq_reqbuf)
888 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
889 req->rq_reqbuf = NULL;
890 req->rq_reqbuf_len = 0;
894 * Given an import \a imp, check whether current user has a valid context
895 * or not. We may create a new context and try to refresh it, and try
896 * repeatedly try in case of non-fatal errors. Return 0 means success.
898 int sptlrpc_import_check_ctx(struct obd_import *imp)
900 struct ptlrpc_sec *sec;
901 struct ptlrpc_cli_ctx *ctx;
902 struct ptlrpc_request *req = NULL;
908 sec = sptlrpc_import_sec_ref(imp);
909 ctx = get_my_ctx(sec);
910 sptlrpc_sec_put(sec);
915 if (cli_ctx_is_eternal(ctx) ||
916 ctx->cc_ops->validate(ctx) == 0) {
917 sptlrpc_cli_ctx_put(ctx, 1);
921 if (cli_ctx_is_error(ctx)) {
922 sptlrpc_cli_ctx_put(ctx, 1);
926 req = ptlrpc_request_cache_alloc(GFP_NOFS);
930 spin_lock_init(&req->rq_lock);
931 atomic_set(&req->rq_refcount, 10000);
932 INIT_LIST_HEAD(&req->rq_ctx_chain);
933 init_waitqueue_head(&req->rq_reply_waitq);
934 init_waitqueue_head(&req->rq_set_waitq);
935 req->rq_import = imp;
936 req->rq_flvr = sec->ps_flvr;
937 req->rq_cli_ctx = ctx;
939 rc = sptlrpc_req_refresh_ctx(req, 0);
940 LASSERT(list_empty(&req->rq_ctx_chain));
941 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
942 ptlrpc_request_cache_free(req);
948 * Used by ptlrpc client, to perform the pre-defined security transformation
949 * upon the request message of \a req. After this function called,
950 * req->rq_reqmsg is still accessible as clear text.
952 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
954 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
959 LASSERT(ctx->cc_sec);
960 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
962 /* we wrap bulk request here because now we can be sure
963 * the context is uptodate.
966 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
971 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
972 case SPTLRPC_SVC_NULL:
973 case SPTLRPC_SVC_AUTH:
974 case SPTLRPC_SVC_INTG:
975 LASSERT(ctx->cc_ops->sign);
976 rc = ctx->cc_ops->sign(ctx, req);
978 case SPTLRPC_SVC_PRIV:
979 LASSERT(ctx->cc_ops->seal);
980 rc = ctx->cc_ops->seal(ctx, req);
987 LASSERT(req->rq_reqdata_len);
988 LASSERT(req->rq_reqdata_len % 8 == 0);
989 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
995 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
997 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1002 LASSERT(ctx->cc_sec);
1003 LASSERT(req->rq_repbuf);
1004 LASSERT(req->rq_repdata);
1005 LASSERT(req->rq_repmsg == NULL);
1007 req->rq_rep_swab_mask = 0;
1009 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1012 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1016 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1020 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1021 CERROR("replied data length %d too small\n",
1022 req->rq_repdata_len);
1026 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1027 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1028 CERROR("reply policy %u doesn't match request policy %u\n",
1029 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1030 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1034 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1035 case SPTLRPC_SVC_NULL:
1036 case SPTLRPC_SVC_AUTH:
1037 case SPTLRPC_SVC_INTG:
1038 LASSERT(ctx->cc_ops->verify);
1039 rc = ctx->cc_ops->verify(ctx, req);
1041 case SPTLRPC_SVC_PRIV:
1042 LASSERT(ctx->cc_ops->unseal);
1043 rc = ctx->cc_ops->unseal(ctx, req);
1048 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1050 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1052 req->rq_rep_swab_mask = 0;
1057 * Used by ptlrpc client, to perform security transformation upon the reply
1058 * message of \a req. After return successfully, req->rq_repmsg points to
1059 * the reply message in clear text.
1061 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1064 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1066 LASSERT(req->rq_repbuf);
1067 LASSERT(req->rq_repdata == NULL);
1068 LASSERT(req->rq_repmsg == NULL);
1069 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1071 if (req->rq_reply_off == 0 &&
1072 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1073 CERROR("real reply with offset 0\n");
1077 if (req->rq_reply_off % 8 != 0) {
1078 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1082 req->rq_repdata = (struct lustre_msg *)
1083 (req->rq_repbuf + req->rq_reply_off);
1084 req->rq_repdata_len = req->rq_nob_received;
1086 return do_cli_unwrap_reply(req);
1090 * Used by ptlrpc client, to perform security transformation upon the early
1091 * reply message of \a req. We expect the rq_reply_off is 0, and
1092 * rq_nob_received is the early reply size.
1094 * Because the receive buffer might be still posted, the reply data might be
1095 * changed at any time, no matter we're holding rq_lock or not. For this reason
1096 * we allocate a separate ptlrpc_request and reply buffer for early reply
1099 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1100 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1101 * \a *req_ret to release it.
1102 * \retval -ev error number, and \a req_ret will not be set.
1104 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1105 struct ptlrpc_request **req_ret)
1107 struct ptlrpc_request *early_req;
1109 int early_bufsz, early_size;
1113 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1114 if (early_req == NULL)
1117 early_size = req->rq_nob_received;
1118 early_bufsz = size_roundup_power2(early_size);
1119 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1120 if (early_buf == NULL)
1121 GOTO(err_req, rc = -ENOMEM);
1123 /* sanity checkings and copy data out, do it inside spinlock */
1124 spin_lock(&req->rq_lock);
1126 if (req->rq_replied) {
1127 spin_unlock(&req->rq_lock);
1128 GOTO(err_buf, rc = -EALREADY);
1131 LASSERT(req->rq_repbuf);
1132 LASSERT(req->rq_repdata == NULL);
1133 LASSERT(req->rq_repmsg == NULL);
1135 if (req->rq_reply_off != 0) {
1136 CERROR("early reply with offset %u\n", req->rq_reply_off);
1137 spin_unlock(&req->rq_lock);
1138 GOTO(err_buf, rc = -EPROTO);
1141 if (req->rq_nob_received != early_size) {
1142 /* even another early arrived the size should be the same */
1143 CERROR("data size has changed from %u to %u\n",
1144 early_size, req->rq_nob_received);
1145 spin_unlock(&req->rq_lock);
1146 GOTO(err_buf, rc = -EINVAL);
1149 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1150 CERROR("early reply length %d too small\n",
1151 req->rq_nob_received);
1152 spin_unlock(&req->rq_lock);
1153 GOTO(err_buf, rc = -EALREADY);
1156 memcpy(early_buf, req->rq_repbuf, early_size);
1157 spin_unlock(&req->rq_lock);
1159 spin_lock_init(&early_req->rq_lock);
1160 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1161 early_req->rq_flvr = req->rq_flvr;
1162 early_req->rq_repbuf = early_buf;
1163 early_req->rq_repbuf_len = early_bufsz;
1164 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1165 early_req->rq_repdata_len = early_size;
1166 early_req->rq_early = 1;
1167 early_req->rq_reqmsg = req->rq_reqmsg;
1169 rc = do_cli_unwrap_reply(early_req);
1171 DEBUG_REQ(D_ADAPTTO, early_req,
1172 "error %d unwrap early reply", rc);
1176 LASSERT(early_req->rq_repmsg);
1177 *req_ret = early_req;
1181 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1183 OBD_FREE_LARGE(early_buf, early_bufsz);
1185 ptlrpc_request_cache_free(early_req);
1190 * Used by ptlrpc client, to release a processed early reply \a early_req.
1192 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1194 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1196 LASSERT(early_req->rq_repbuf);
1197 LASSERT(early_req->rq_repdata);
1198 LASSERT(early_req->rq_repmsg);
1200 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1201 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1202 ptlrpc_request_cache_free(early_req);
1205 /**************************************************
1207 **************************************************/
1210 * "fixed" sec (e.g. null) use sec_id < 0
1212 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1214 int sptlrpc_get_next_secid(void)
1216 return atomic_inc_return(&sptlrpc_sec_id);
1218 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1220 /**************************************************
1221 * client side high-level security APIs *
1222 **************************************************/
1224 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1225 int grace, int force)
1227 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1229 LASSERT(policy->sp_cops);
1230 LASSERT(policy->sp_cops->flush_ctx_cache);
1232 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1235 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1237 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1239 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1240 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1241 LASSERT(policy->sp_cops->destroy_sec);
1243 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1245 policy->sp_cops->destroy_sec(sec);
1246 sptlrpc_policy_put(policy);
1249 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1251 sec_cop_destroy_sec(sec);
1253 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1255 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1257 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1259 if (sec->ps_policy->sp_cops->kill_sec) {
1260 sec->ps_policy->sp_cops->kill_sec(sec);
1262 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1266 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1269 atomic_inc(&sec->ps_refcount);
1273 EXPORT_SYMBOL(sptlrpc_sec_get);
1275 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1278 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1280 if (atomic_dec_and_test(&sec->ps_refcount)) {
1281 sptlrpc_gc_del_sec(sec);
1282 sec_cop_destroy_sec(sec);
1286 EXPORT_SYMBOL(sptlrpc_sec_put);
1289 * policy module is responsible for taking refrence of import
1292 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1293 struct ptlrpc_svc_ctx *svc_ctx,
1294 struct sptlrpc_flavor *sf,
1295 enum lustre_sec_part sp)
1297 struct ptlrpc_sec_policy *policy;
1298 struct ptlrpc_sec *sec;
1303 LASSERT(imp->imp_dlm_fake == 1);
1305 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1306 imp->imp_obd->obd_type->typ_name,
1307 imp->imp_obd->obd_name,
1308 sptlrpc_flavor2name(sf, str, sizeof(str)));
1310 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1311 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1313 LASSERT(imp->imp_dlm_fake == 0);
1315 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1316 imp->imp_obd->obd_type->typ_name,
1317 imp->imp_obd->obd_name,
1318 sptlrpc_flavor2name(sf, str, sizeof(str)));
1320 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1322 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1327 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1329 atomic_inc(&sec->ps_refcount);
1333 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1334 sptlrpc_gc_add_sec(sec);
1336 sptlrpc_policy_put(policy);
1342 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1344 struct ptlrpc_sec *sec;
1346 spin_lock(&imp->imp_lock);
1347 sec = sptlrpc_sec_get(imp->imp_sec);
1348 spin_unlock(&imp->imp_lock);
1352 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1354 static void sptlrpc_import_sec_install(struct obd_import *imp,
1355 struct ptlrpc_sec *sec)
1357 struct ptlrpc_sec *old_sec;
1359 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1361 spin_lock(&imp->imp_lock);
1362 old_sec = imp->imp_sec;
1364 spin_unlock(&imp->imp_lock);
1367 sptlrpc_sec_kill(old_sec);
1369 /* balance the ref taken by this import */
1370 sptlrpc_sec_put(old_sec);
1375 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1377 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1381 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1386 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1387 struct ptlrpc_sec *sec,
1388 struct sptlrpc_flavor *sf)
1390 char str1[32], str2[32];
1392 if (sec->ps_flvr.sf_flags != sf->sf_flags)
1393 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1394 sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1395 str1, sizeof(str1)),
1396 sptlrpc_secflags2str(sf->sf_flags,
1397 str2, sizeof(str2)));
1399 spin_lock(&sec->ps_lock);
1400 flavor_copy(&sec->ps_flvr, sf);
1401 spin_unlock(&sec->ps_lock);
1405 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1406 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1408 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1409 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1411 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1412 struct ptlrpc_svc_ctx *svc_ctx,
1413 struct sptlrpc_flavor *flvr)
1415 struct ptlrpc_connection *conn;
1416 struct sptlrpc_flavor sf;
1417 struct ptlrpc_sec *sec, *newsec;
1418 enum lustre_sec_part sp;
1428 conn = imp->imp_connection;
1430 if (svc_ctx == NULL) {
1431 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1433 * normal import, determine flavor from rule set, except
1434 * for mgc the flavor is predetermined.
1436 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1437 sf = cliobd->cl_flvr_mgc;
1439 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1441 &cliobd->cl_target_uuid,
1444 sp = imp->imp_obd->u.cli.cl_sp_me;
1446 /* reverse import, determine flavor from incoming reqeust */
1449 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1450 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1451 PTLRPC_SEC_FL_ROOTONLY;
1453 sp = sptlrpc_target_sec_part(imp->imp_obd);
1456 sec = sptlrpc_import_sec_ref(imp);
1460 if (flavor_equal(&sf, &sec->ps_flvr))
1463 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1464 imp->imp_obd->obd_name,
1465 obd_uuid2str(&conn->c_remote_uuid),
1466 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1467 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1469 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1470 SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1471 SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1472 SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1473 sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1476 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1477 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1478 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1479 imp->imp_obd->obd_name,
1480 obd_uuid2str(&conn->c_remote_uuid),
1481 LNET_NIDNET(conn->c_self),
1482 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1485 mutex_lock(&imp->imp_sec_mutex);
1487 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1489 sptlrpc_import_sec_install(imp, newsec);
1491 CERROR("import %s->%s: failed to create new sec\n",
1492 imp->imp_obd->obd_name,
1493 obd_uuid2str(&conn->c_remote_uuid));
1497 mutex_unlock(&imp->imp_sec_mutex);
1499 sptlrpc_sec_put(sec);
1503 void sptlrpc_import_sec_put(struct obd_import *imp)
1506 sptlrpc_sec_kill(imp->imp_sec);
1508 sptlrpc_sec_put(imp->imp_sec);
1509 imp->imp_sec = NULL;
1513 static void import_flush_ctx_common(struct obd_import *imp,
1514 uid_t uid, int grace, int force)
1516 struct ptlrpc_sec *sec;
1521 sec = sptlrpc_import_sec_ref(imp);
1525 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1526 sptlrpc_sec_put(sec);
1529 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1531 /* it's important to use grace mode, see explain in
1532 * sptlrpc_req_refresh_ctx() */
1533 import_flush_ctx_common(imp, 0, 1, 1);
1536 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1538 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1541 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1543 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1545 import_flush_ctx_common(imp, -1, 1, 1);
1547 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1550 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1551 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1553 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1555 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1556 struct ptlrpc_sec_policy *policy;
1560 LASSERT(ctx->cc_sec);
1561 LASSERT(ctx->cc_sec->ps_policy);
1562 LASSERT(req->rq_reqmsg == NULL);
1563 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1565 policy = ctx->cc_sec->ps_policy;
1566 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1568 LASSERT(req->rq_reqmsg);
1569 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1571 /* zeroing preallocated buffer */
1573 memset(req->rq_reqmsg, 0, msgsize);
1580 * Used by ptlrpc client to free request buffer of \a req. After this
1581 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1583 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1585 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1586 struct ptlrpc_sec_policy *policy;
1589 LASSERT(ctx->cc_sec);
1590 LASSERT(ctx->cc_sec->ps_policy);
1591 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1593 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1596 policy = ctx->cc_sec->ps_policy;
1597 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1598 req->rq_reqmsg = NULL;
1602 * NOTE caller must guarantee the buffer size is enough for the enlargement
1604 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1605 int segment, int newsize)
1608 int oldsize, oldmsg_size, movesize;
1610 LASSERT(segment < msg->lm_bufcount);
1611 LASSERT(msg->lm_buflens[segment] <= newsize);
1613 if (msg->lm_buflens[segment] == newsize)
1616 /* nothing to do if we are enlarging the last segment */
1617 if (segment == msg->lm_bufcount - 1) {
1618 msg->lm_buflens[segment] = newsize;
1622 oldsize = msg->lm_buflens[segment];
1624 src = lustre_msg_buf(msg, segment + 1, 0);
1625 msg->lm_buflens[segment] = newsize;
1626 dst = lustre_msg_buf(msg, segment + 1, 0);
1627 msg->lm_buflens[segment] = oldsize;
1629 /* move from segment + 1 to end segment */
1630 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1631 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1632 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1633 LASSERT(movesize >= 0);
1636 memmove(dst, src, movesize);
1638 /* note we don't clear the ares where old data live, not secret */
1640 /* finally set new segment size */
1641 msg->lm_buflens[segment] = newsize;
1643 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1646 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1647 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1648 * preserved after the enlargement. this must be called after original request
1649 * buffer being allocated.
1651 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1652 * so caller should refresh its local pointers if needed.
1654 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1655 int segment, int newsize)
1657 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1658 struct ptlrpc_sec_cops *cops;
1659 struct lustre_msg *msg = req->rq_reqmsg;
1663 LASSERT(msg->lm_bufcount > segment);
1664 LASSERT(msg->lm_buflens[segment] <= newsize);
1666 if (msg->lm_buflens[segment] == newsize)
1669 cops = ctx->cc_sec->ps_policy->sp_cops;
1670 LASSERT(cops->enlarge_reqbuf);
1671 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1673 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1676 * Used by ptlrpc client to allocate reply buffer of \a req.
1678 * \note After this, req->rq_repmsg is still not accessible.
1680 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1682 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1683 struct ptlrpc_sec_policy *policy;
1687 LASSERT(ctx->cc_sec);
1688 LASSERT(ctx->cc_sec->ps_policy);
1693 policy = ctx->cc_sec->ps_policy;
1694 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1698 * Used by ptlrpc client to free reply buffer of \a req. After this
1699 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1701 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1703 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1704 struct ptlrpc_sec_policy *policy;
1708 LASSERT(ctx->cc_sec);
1709 LASSERT(ctx->cc_sec->ps_policy);
1710 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1712 if (req->rq_repbuf == NULL)
1714 LASSERT(req->rq_repbuf_len);
1716 policy = ctx->cc_sec->ps_policy;
1717 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1718 req->rq_repmsg = NULL;
1722 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1723 struct ptlrpc_cli_ctx *ctx)
1725 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1727 if (!policy->sp_cops->install_rctx)
1729 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1732 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1733 struct ptlrpc_svc_ctx *ctx)
1735 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1737 if (!policy->sp_sops->install_rctx)
1739 return policy->sp_sops->install_rctx(imp, ctx);
1742 /****************************************
1743 * server side security *
1744 ****************************************/
1746 static int flavor_allowed(struct sptlrpc_flavor *exp,
1747 struct ptlrpc_request *req)
1749 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1751 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1754 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1755 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1756 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1757 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1763 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1766 * Given an export \a exp, check whether the flavor of incoming \a req
1767 * is allowed by the export \a exp. Main logic is about taking care of
1768 * changing configurations. Return 0 means success.
1770 int sptlrpc_target_export_check(struct obd_export *exp,
1771 struct ptlrpc_request *req)
1773 struct sptlrpc_flavor flavor;
1778 /* client side export has no imp_reverse, skip
1779 * FIXME maybe we should check flavor this as well??? */
1780 if (exp->exp_imp_reverse == NULL)
1783 /* don't care about ctx fini rpc */
1784 if (req->rq_ctx_fini)
1787 spin_lock(&exp->exp_lock);
1789 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1790 * the first req with the new flavor, then treat it as current flavor,
1791 * adapt reverse sec according to it.
1792 * note the first rpc with new flavor might not be with root ctx, in
1793 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1794 if (unlikely(exp->exp_flvr_changed) &&
1795 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1796 /* make the new flavor as "current", and old ones as
1797 * about-to-expire */
1798 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1799 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1800 flavor = exp->exp_flvr_old[1];
1801 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1802 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1803 exp->exp_flvr_old[0] = exp->exp_flvr;
1804 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1805 EXP_FLVR_UPDATE_EXPIRE;
1806 exp->exp_flvr = flavor;
1808 /* flavor change finished */
1809 exp->exp_flvr_changed = 0;
1810 LASSERT(exp->exp_flvr_adapt == 1);
1812 /* if it's gss, we only interested in root ctx init */
1813 if (req->rq_auth_gss &&
1814 !(req->rq_ctx_init &&
1815 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1816 req->rq_auth_usr_ost))) {
1817 spin_unlock(&exp->exp_lock);
1818 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1819 req->rq_auth_gss, req->rq_ctx_init,
1820 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1821 req->rq_auth_usr_ost);
1825 exp->exp_flvr_adapt = 0;
1826 spin_unlock(&exp->exp_lock);
1828 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1829 req->rq_svc_ctx, &flavor);
1832 /* if it equals to the current flavor, we accept it, but need to
1833 * dealing with reverse sec/ctx */
1834 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1835 /* most cases should return here, we only interested in
1836 * gss root ctx init */
1837 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1838 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1839 !req->rq_auth_usr_ost)) {
1840 spin_unlock(&exp->exp_lock);
1844 /* if flavor just changed, we should not proceed, just leave
1845 * it and current flavor will be discovered and replaced
1846 * shortly, and let _this_ rpc pass through */
1847 if (exp->exp_flvr_changed) {
1848 LASSERT(exp->exp_flvr_adapt);
1849 spin_unlock(&exp->exp_lock);
1853 if (exp->exp_flvr_adapt) {
1854 exp->exp_flvr_adapt = 0;
1855 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1856 exp, exp->exp_flvr.sf_rpc,
1857 exp->exp_flvr_old[0].sf_rpc,
1858 exp->exp_flvr_old[1].sf_rpc);
1859 flavor = exp->exp_flvr;
1860 spin_unlock(&exp->exp_lock);
1862 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1866 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1867 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1868 exp->exp_flvr_old[0].sf_rpc,
1869 exp->exp_flvr_old[1].sf_rpc);
1870 spin_unlock(&exp->exp_lock);
1872 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1877 if (exp->exp_flvr_expire[0]) {
1878 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1879 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1880 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1881 "middle one ("CFS_DURATION_T")\n", exp,
1882 exp->exp_flvr.sf_rpc,
1883 exp->exp_flvr_old[0].sf_rpc,
1884 exp->exp_flvr_old[1].sf_rpc,
1885 exp->exp_flvr_expire[0] -
1886 cfs_time_current_sec());
1887 spin_unlock(&exp->exp_lock);
1891 CDEBUG(D_SEC, "mark middle expired\n");
1892 exp->exp_flvr_expire[0] = 0;
1894 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1895 exp->exp_flvr.sf_rpc,
1896 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1897 req->rq_flvr.sf_rpc);
1900 /* now it doesn't match the current flavor, the only chance we can
1901 * accept it is match the old flavors which is not expired. */
1902 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1903 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1904 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1905 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1906 "oldest one ("CFS_DURATION_T")\n", exp,
1907 exp->exp_flvr.sf_rpc,
1908 exp->exp_flvr_old[0].sf_rpc,
1909 exp->exp_flvr_old[1].sf_rpc,
1910 exp->exp_flvr_expire[1] -
1911 cfs_time_current_sec());
1912 spin_unlock(&exp->exp_lock);
1916 CDEBUG(D_SEC, "mark oldest expired\n");
1917 exp->exp_flvr_expire[1] = 0;
1919 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1920 exp, exp->exp_flvr.sf_rpc,
1921 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1922 req->rq_flvr.sf_rpc);
1924 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1925 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1926 exp->exp_flvr_old[1].sf_rpc);
1929 spin_unlock(&exp->exp_lock);
1931 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1932 "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1933 exp, exp->exp_obd->obd_name,
1934 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1935 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1936 req->rq_flvr.sf_rpc,
1937 exp->exp_flvr.sf_rpc,
1938 exp->exp_flvr_old[0].sf_rpc,
1939 exp->exp_flvr_expire[0] ?
1940 (unsigned long) (exp->exp_flvr_expire[0] -
1941 cfs_time_current_sec()) : 0,
1942 exp->exp_flvr_old[1].sf_rpc,
1943 exp->exp_flvr_expire[1] ?
1944 (unsigned long) (exp->exp_flvr_expire[1] -
1945 cfs_time_current_sec()) : 0);
1948 EXPORT_SYMBOL(sptlrpc_target_export_check);
1950 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1951 struct sptlrpc_rule_set *rset)
1953 struct obd_export *exp;
1954 struct sptlrpc_flavor new_flvr;
1958 spin_lock(&obd->obd_dev_lock);
1960 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1961 if (exp->exp_connection == NULL)
1964 /* note if this export had just been updated flavor
1965 * (exp_flvr_changed == 1), this will override the
1967 spin_lock(&exp->exp_lock);
1968 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1969 exp->exp_connection->c_peer.nid,
1971 if (exp->exp_flvr_changed ||
1972 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1973 exp->exp_flvr_old[1] = new_flvr;
1974 exp->exp_flvr_expire[1] = 0;
1975 exp->exp_flvr_changed = 1;
1976 exp->exp_flvr_adapt = 1;
1978 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1979 exp, sptlrpc_part2name(exp->exp_sp_peer),
1980 exp->exp_flvr.sf_rpc,
1981 exp->exp_flvr_old[1].sf_rpc);
1983 spin_unlock(&exp->exp_lock);
1986 spin_unlock(&obd->obd_dev_lock);
1988 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1990 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1992 /* peer's claim is unreliable unless gss is being used */
1993 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1996 switch (req->rq_sp_from) {
1998 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
1999 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2000 svc_rc = SECSVC_DROP;
2004 if (!req->rq_auth_usr_mdt) {
2005 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2006 svc_rc = SECSVC_DROP;
2010 if (!req->rq_auth_usr_ost) {
2011 DEBUG_REQ(D_ERROR, req, "faked source OST");
2012 svc_rc = SECSVC_DROP;
2017 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2018 !req->rq_auth_usr_ost) {
2019 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2020 svc_rc = SECSVC_DROP;
2025 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2026 svc_rc = SECSVC_DROP;
2033 * Used by ptlrpc server, to perform transformation upon request message of
2034 * incoming \a req. This must be the first thing to do with a incoming
2035 * request in ptlrpc layer.
2037 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2038 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2039 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2040 * reply message has been prepared.
2041 * \retval SECSVC_DROP failed, this request should be dropped.
2043 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2045 struct ptlrpc_sec_policy *policy;
2046 struct lustre_msg *msg = req->rq_reqbuf;
2051 LASSERT(req->rq_reqmsg == NULL);
2052 LASSERT(req->rq_repmsg == NULL);
2053 LASSERT(req->rq_svc_ctx == NULL);
2055 req->rq_req_swab_mask = 0;
2057 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2060 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2064 CERROR("error unpacking request from %s x"LPU64"\n",
2065 libcfs_id2str(req->rq_peer), req->rq_xid);
2066 RETURN(SECSVC_DROP);
2069 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2070 req->rq_sp_from = LUSTRE_SP_ANY;
2071 req->rq_auth_uid = -1; /* set to INVALID_UID */
2072 req->rq_auth_mapped_uid = -1;
2074 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2076 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2077 RETURN(SECSVC_DROP);
2080 LASSERT(policy->sp_sops->accept);
2081 rc = policy->sp_sops->accept(req);
2082 sptlrpc_policy_put(policy);
2083 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2084 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2087 * if it's not null flavor (which means embedded packing msg),
2088 * reset the swab mask for the comming inner msg unpacking.
2090 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2091 req->rq_req_swab_mask = 0;
2093 /* sanity check for the request source */
2094 rc = sptlrpc_svc_check_from(req, rc);
2099 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2100 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2101 * a buffer of \a msglen size.
2103 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2105 struct ptlrpc_sec_policy *policy;
2106 struct ptlrpc_reply_state *rs;
2110 LASSERT(req->rq_svc_ctx);
2111 LASSERT(req->rq_svc_ctx->sc_policy);
2113 policy = req->rq_svc_ctx->sc_policy;
2114 LASSERT(policy->sp_sops->alloc_rs);
2116 rc = policy->sp_sops->alloc_rs(req, msglen);
2117 if (unlikely(rc == -ENOMEM)) {
2118 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2119 if (svcpt->scp_service->srv_max_reply_size <
2120 msglen + sizeof(struct ptlrpc_reply_state)) {
2121 /* Just return failure if the size is too big */
2122 CERROR("size of message is too big (%zd), %d allowed\n",
2123 msglen + sizeof(struct ptlrpc_reply_state),
2124 svcpt->scp_service->srv_max_reply_size);
2128 /* failed alloc, try emergency pool */
2129 rs = lustre_get_emerg_rs(svcpt);
2133 req->rq_reply_state = rs;
2134 rc = policy->sp_sops->alloc_rs(req, msglen);
2136 lustre_put_emerg_rs(rs);
2137 req->rq_reply_state = NULL;
2142 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2148 * Used by ptlrpc server, to perform transformation upon reply message.
2150 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2151 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2153 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2155 struct ptlrpc_sec_policy *policy;
2159 LASSERT(req->rq_svc_ctx);
2160 LASSERT(req->rq_svc_ctx->sc_policy);
2162 policy = req->rq_svc_ctx->sc_policy;
2163 LASSERT(policy->sp_sops->authorize);
2165 rc = policy->sp_sops->authorize(req);
2166 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2172 * Used by ptlrpc server, to free reply_state.
2174 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2176 struct ptlrpc_sec_policy *policy;
2177 unsigned int prealloc;
2180 LASSERT(rs->rs_svc_ctx);
2181 LASSERT(rs->rs_svc_ctx->sc_policy);
2183 policy = rs->rs_svc_ctx->sc_policy;
2184 LASSERT(policy->sp_sops->free_rs);
2186 prealloc = rs->rs_prealloc;
2187 policy->sp_sops->free_rs(rs);
2190 lustre_put_emerg_rs(rs);
2194 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2196 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2199 atomic_inc(&ctx->sc_refcount);
2202 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2204 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2209 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2210 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2211 if (ctx->sc_policy->sp_sops->free_ctx)
2212 ctx->sc_policy->sp_sops->free_ctx(ctx);
2214 req->rq_svc_ctx = NULL;
2217 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2219 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2224 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2225 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2226 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2228 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2230 /****************************************
2232 ****************************************/
2235 * Perform transformation upon bulk data pointed by \a desc. This is called
2236 * before transforming the request message.
2238 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2239 struct ptlrpc_bulk_desc *desc)
2241 struct ptlrpc_cli_ctx *ctx;
2243 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2245 if (!req->rq_pack_bulk)
2248 ctx = req->rq_cli_ctx;
2249 if (ctx->cc_ops->wrap_bulk)
2250 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2253 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2256 * This is called after unwrap the reply message.
2257 * return nob of actual plain text size received, or error code.
2259 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2260 struct ptlrpc_bulk_desc *desc,
2263 struct ptlrpc_cli_ctx *ctx;
2266 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2268 if (!req->rq_pack_bulk)
2269 return desc->bd_nob_transferred;
2271 ctx = req->rq_cli_ctx;
2272 if (ctx->cc_ops->unwrap_bulk) {
2273 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2277 return desc->bd_nob_transferred;
2279 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2282 * This is called after unwrap the reply message.
2283 * return 0 for success or error code.
2285 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2286 struct ptlrpc_bulk_desc *desc)
2288 struct ptlrpc_cli_ctx *ctx;
2291 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2293 if (!req->rq_pack_bulk)
2296 ctx = req->rq_cli_ctx;
2297 if (ctx->cc_ops->unwrap_bulk) {
2298 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2304 * if everything is going right, nob should equals to nob_transferred.
2305 * in case of privacy mode, nob_transferred needs to be adjusted.
2307 if (desc->bd_nob != desc->bd_nob_transferred) {
2308 CERROR("nob %d doesn't match transferred nob %d\n",
2309 desc->bd_nob, desc->bd_nob_transferred);
2315 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2317 #ifdef HAVE_SERVER_SUPPORT
2319 * Performe transformation upon outgoing bulk read.
2321 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2322 struct ptlrpc_bulk_desc *desc)
2324 struct ptlrpc_svc_ctx *ctx;
2326 LASSERT(req->rq_bulk_read);
2328 if (!req->rq_pack_bulk)
2331 ctx = req->rq_svc_ctx;
2332 if (ctx->sc_policy->sp_sops->wrap_bulk)
2333 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2337 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2340 * Performe transformation upon incoming bulk write.
2342 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2343 struct ptlrpc_bulk_desc *desc)
2345 struct ptlrpc_svc_ctx *ctx;
2348 LASSERT(req->rq_bulk_write);
2351 * if it's in privacy mode, transferred should >= expected; otherwise
2352 * transferred should == expected.
2354 if (desc->bd_nob_transferred < desc->bd_nob ||
2355 (desc->bd_nob_transferred > desc->bd_nob &&
2356 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2357 SPTLRPC_BULK_SVC_PRIV)) {
2358 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2359 desc->bd_nob_transferred, desc->bd_nob);
2363 if (!req->rq_pack_bulk)
2366 ctx = req->rq_svc_ctx;
2367 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2368 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2370 CERROR("error unwrap bulk: %d\n", rc);
2373 /* return 0 to allow reply be sent */
2376 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2379 * Prepare buffers for incoming bulk write.
2381 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2382 struct ptlrpc_bulk_desc *desc)
2384 struct ptlrpc_svc_ctx *ctx;
2386 LASSERT(req->rq_bulk_write);
2388 if (!req->rq_pack_bulk)
2391 ctx = req->rq_svc_ctx;
2392 if (ctx->sc_policy->sp_sops->prep_bulk)
2393 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2397 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2399 #endif /* HAVE_SERVER_SUPPORT */
2401 /****************************************
2402 * user descriptor helpers *
2403 ****************************************/
2405 int sptlrpc_current_user_desc_size(void)
2409 ngroups = current_ngroups;
2411 if (ngroups > LUSTRE_MAX_GROUPS)
2412 ngroups = LUSTRE_MAX_GROUPS;
2413 return sptlrpc_user_desc_size(ngroups);
2415 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2417 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2419 struct ptlrpc_user_desc *pud;
2421 pud = lustre_msg_buf(msg, offset, 0);
2423 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2424 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2425 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2426 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2427 pud->pud_cap = cfs_curproc_cap_pack();
2428 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2431 if (pud->pud_ngroups > current_ngroups)
2432 pud->pud_ngroups = current_ngroups;
2433 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2434 pud->pud_ngroups * sizeof(__u32));
2435 task_unlock(current);
2439 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2441 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2443 struct ptlrpc_user_desc *pud;
2446 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2451 __swab32s(&pud->pud_uid);
2452 __swab32s(&pud->pud_gid);
2453 __swab32s(&pud->pud_fsuid);
2454 __swab32s(&pud->pud_fsgid);
2455 __swab32s(&pud->pud_cap);
2456 __swab32s(&pud->pud_ngroups);
2459 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2460 CERROR("%u groups is too large\n", pud->pud_ngroups);
2464 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2465 msg->lm_buflens[offset]) {
2466 CERROR("%u groups are claimed but bufsize only %u\n",
2467 pud->pud_ngroups, msg->lm_buflens[offset]);
2472 for (i = 0; i < pud->pud_ngroups; i++)
2473 __swab32s(&pud->pud_groups[i]);
2478 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2480 /****************************************
2482 ****************************************/
2484 const char * sec2target_str(struct ptlrpc_sec *sec)
2486 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2488 if (sec_is_reverse(sec))
2490 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2492 EXPORT_SYMBOL(sec2target_str);
2495 * return true if the bulk data is protected
2497 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2499 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2500 case SPTLRPC_BULK_SVC_INTG:
2501 case SPTLRPC_BULK_SVC_PRIV:
2507 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2509 /****************************************
2510 * crypto API helper/alloc blkciper *
2511 ****************************************/
2513 /****************************************
2514 * initialize/finalize *
2515 ****************************************/
2517 int sptlrpc_init(void)
2521 rwlock_init(&policy_lock);
2523 rc = sptlrpc_gc_init();
2527 rc = sptlrpc_conf_init();
2531 rc = sptlrpc_enc_pool_init();
2535 rc = sptlrpc_null_init();
2539 rc = sptlrpc_plain_init();
2543 rc = sptlrpc_lproc_init();
2550 sptlrpc_plain_fini();
2552 sptlrpc_null_fini();
2554 sptlrpc_enc_pool_fini();
2556 sptlrpc_conf_fini();
2563 void sptlrpc_fini(void)
2565 sptlrpc_lproc_fini();
2566 sptlrpc_plain_fini();
2567 sptlrpc_null_fini();
2568 sptlrpc_enc_pool_fini();
2569 sptlrpc_conf_fini();