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, 2014, 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 <linux/user_namespace.h>
44 #ifdef HAVE_UIDGID_HEADER
45 # include <linux/uidgid.h>
47 #include <linux/crypto.h>
48 #include <linux/key.h>
50 #include <libcfs/libcfs.h>
52 #include <obd_class.h>
53 #include <obd_support.h>
54 #include <lustre_net.h>
55 #include <lustre_import.h>
56 #include <lustre_dlm.h>
57 #include <lustre_sec.h>
59 #include "ptlrpc_internal.h"
61 /***********************************************
63 ***********************************************/
65 static rwlock_t policy_lock;
66 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
70 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
72 __u16 number = policy->sp_policy;
74 LASSERT(policy->sp_name);
75 LASSERT(policy->sp_cops);
76 LASSERT(policy->sp_sops);
78 if (number >= SPTLRPC_POLICY_MAX)
81 write_lock(&policy_lock);
82 if (unlikely(policies[number])) {
83 write_unlock(&policy_lock);
86 policies[number] = policy;
87 write_unlock(&policy_lock);
89 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
92 EXPORT_SYMBOL(sptlrpc_register_policy);
94 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
96 __u16 number = policy->sp_policy;
98 LASSERT(number < SPTLRPC_POLICY_MAX);
100 write_lock(&policy_lock);
101 if (unlikely(policies[number] == NULL)) {
102 write_unlock(&policy_lock);
103 CERROR("%s: already unregistered\n", policy->sp_name);
107 LASSERT(policies[number] == policy);
108 policies[number] = NULL;
109 write_unlock(&policy_lock);
111 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
114 EXPORT_SYMBOL(sptlrpc_unregister_policy);
117 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
119 static DEFINE_MUTEX(load_mutex);
120 static atomic_t loaded = ATOMIC_INIT(0);
121 struct ptlrpc_sec_policy *policy;
122 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
125 if (number >= SPTLRPC_POLICY_MAX)
129 read_lock(&policy_lock);
130 policy = policies[number];
131 if (policy && !try_module_get(policy->sp_owner))
134 flag = atomic_read(&loaded);
135 read_unlock(&policy_lock);
137 if (policy != NULL || flag != 0 ||
138 number != SPTLRPC_POLICY_GSS)
141 /* try to load gss module, once */
142 mutex_lock(&load_mutex);
143 if (atomic_read(&loaded) == 0) {
144 if (request_module("ptlrpc_gss") == 0)
146 "module ptlrpc_gss loaded on demand\n");
148 CERROR("Unable to load module ptlrpc_gss\n");
150 atomic_set(&loaded, 1);
152 mutex_unlock(&load_mutex);
158 __u32 sptlrpc_name2flavor_base(const char *name)
160 if (!strcmp(name, "null"))
161 return SPTLRPC_FLVR_NULL;
162 if (!strcmp(name, "plain"))
163 return SPTLRPC_FLVR_PLAIN;
164 if (!strcmp(name, "gssnull"))
165 return SPTLRPC_FLVR_GSSNULL;
166 if (!strcmp(name, "krb5n"))
167 return SPTLRPC_FLVR_KRB5N;
168 if (!strcmp(name, "krb5a"))
169 return SPTLRPC_FLVR_KRB5A;
170 if (!strcmp(name, "krb5i"))
171 return SPTLRPC_FLVR_KRB5I;
172 if (!strcmp(name, "krb5p"))
173 return SPTLRPC_FLVR_KRB5P;
174 if (!strcmp(name, "ski"))
175 return SPTLRPC_FLVR_SKI;
176 if (!strcmp(name, "skpi"))
177 return SPTLRPC_FLVR_SKPI;
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_GSSNULL))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
199 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
201 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
203 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
206 CERROR("invalid wire flavor 0x%x\n", flvr);
209 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
211 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
212 char *buf, int bufsize)
214 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
215 snprintf(buf, bufsize, "hash:%s",
216 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
218 snprintf(buf, bufsize, "%s",
219 sptlrpc_flavor2name_base(sf->sf_rpc));
221 buf[bufsize - 1] = '\0';
224 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
226 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
228 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
231 * currently we don't support customized bulk specification for
232 * flavors other than plain
234 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
238 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
239 strncat(buf, bspec, bufsize);
242 buf[bufsize - 1] = '\0';
245 EXPORT_SYMBOL(sptlrpc_flavor2name);
247 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
251 if (flags & PTLRPC_SEC_FL_REVERSE)
252 strlcat(buf, "reverse,", bufsize);
253 if (flags & PTLRPC_SEC_FL_ROOTONLY)
254 strlcat(buf, "rootonly,", bufsize);
255 if (flags & PTLRPC_SEC_FL_UDESC)
256 strlcat(buf, "udesc,", bufsize);
257 if (flags & PTLRPC_SEC_FL_BULK)
258 strlcat(buf, "bulk,", bufsize);
260 strlcat(buf, "-,", bufsize);
264 EXPORT_SYMBOL(sptlrpc_secflags2str);
266 /**************************************************
267 * client context APIs *
268 **************************************************/
271 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
273 struct vfs_cred vcred;
274 int create = 1, remove_dead = 1;
277 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
279 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
280 PTLRPC_SEC_FL_ROOTONLY)) {
283 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
288 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
289 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
292 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
296 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
298 atomic_inc(&ctx->cc_refcount);
301 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
303 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
305 struct ptlrpc_sec *sec = ctx->cc_sec;
308 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
310 if (!atomic_dec_and_test(&ctx->cc_refcount))
313 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
315 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
318 * Expire the client context immediately.
320 * \pre Caller must hold at least 1 reference on the \a ctx.
322 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
324 LASSERT(ctx->cc_ops->die);
325 ctx->cc_ops->die(ctx, 0);
327 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
330 * To wake up the threads who are waiting for this client context. Called
331 * after some status change happened on \a ctx.
333 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
335 struct ptlrpc_request *req, *next;
337 spin_lock(&ctx->cc_lock);
338 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
340 list_del_init(&req->rq_ctx_chain);
341 ptlrpc_client_wake_req(req);
343 spin_unlock(&ctx->cc_lock);
345 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
347 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
349 LASSERT(ctx->cc_ops);
351 if (ctx->cc_ops->display == NULL)
354 return ctx->cc_ops->display(ctx, buf, bufsize);
357 static int import_sec_check_expire(struct obd_import *imp)
361 spin_lock(&imp->imp_lock);
362 if (imp->imp_sec_expire &&
363 imp->imp_sec_expire < cfs_time_current_sec()) {
365 imp->imp_sec_expire = 0;
367 spin_unlock(&imp->imp_lock);
372 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
373 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
377 * Get and validate the client side ptlrpc security facilities from
378 * \a imp. There is a race condition on client reconnect when the import is
379 * being destroyed while there are outstanding client bound requests. In
380 * this case do not output any error messages if import secuity is not
383 * \param[in] imp obd import associated with client
384 * \param[out] sec client side ptlrpc security
386 * \retval 0 if security retrieved successfully
387 * \retval -ve errno if there was a problem
389 static int import_sec_validate_get(struct obd_import *imp,
390 struct ptlrpc_sec **sec)
394 if (unlikely(imp->imp_sec_expire)) {
395 rc = import_sec_check_expire(imp);
400 *sec = sptlrpc_import_sec_ref(imp);
402 CERROR("import %p (%s) with no sec\n",
403 imp, ptlrpc_import_state_name(imp->imp_state));
407 if (unlikely((*sec)->ps_dying)) {
408 CERROR("attempt to use dying sec %p\n", sec);
409 sptlrpc_sec_put(*sec);
417 * Given a \a req, find or allocate an appropriate context for it.
418 * \pre req->rq_cli_ctx == NULL.
420 * \retval 0 succeed, and req->rq_cli_ctx is set.
421 * \retval -ev error number, and req->rq_cli_ctx == NULL.
423 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
425 struct obd_import *imp = req->rq_import;
426 struct ptlrpc_sec *sec;
430 LASSERT(!req->rq_cli_ctx);
433 rc = import_sec_validate_get(imp, &sec);
437 req->rq_cli_ctx = get_my_ctx(sec);
439 sptlrpc_sec_put(sec);
441 if (!req->rq_cli_ctx) {
442 CERROR("req %p: fail to get context\n", req);
443 RETURN(-ECONNREFUSED);
450 * Drop the context for \a req.
451 * \pre req->rq_cli_ctx != NULL.
452 * \post req->rq_cli_ctx == NULL.
454 * If \a sync == 0, this function should return quickly without sleep;
455 * otherwise it might trigger and wait for the whole process of sending
456 * an context-destroying rpc to server.
458 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
463 LASSERT(req->rq_cli_ctx);
465 /* request might be asked to release earlier while still
466 * in the context waiting list.
468 if (!list_empty(&req->rq_ctx_chain)) {
469 spin_lock(&req->rq_cli_ctx->cc_lock);
470 list_del_init(&req->rq_ctx_chain);
471 spin_unlock(&req->rq_cli_ctx->cc_lock);
474 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
475 req->rq_cli_ctx = NULL;
480 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
481 struct ptlrpc_cli_ctx *oldctx,
482 struct ptlrpc_cli_ctx *newctx)
484 struct sptlrpc_flavor old_flvr;
485 char *reqmsg = NULL; /* to workaround old gcc */
489 LASSERT(req->rq_reqmsg);
490 LASSERT(req->rq_reqlen);
491 LASSERT(req->rq_replen);
493 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
494 "switch sec %p(%s) -> %p(%s)\n", req,
495 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
496 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
497 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
498 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
501 old_flvr = req->rq_flvr;
503 /* save request message */
504 reqmsg_size = req->rq_reqlen;
505 if (reqmsg_size != 0) {
506 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
509 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
512 /* release old req/rep buf */
513 req->rq_cli_ctx = oldctx;
514 sptlrpc_cli_free_reqbuf(req);
515 sptlrpc_cli_free_repbuf(req);
516 req->rq_cli_ctx = newctx;
518 /* recalculate the flavor */
519 sptlrpc_req_set_flavor(req, 0);
521 /* alloc new request buffer
522 * we don't need to alloc reply buffer here, leave it to the
523 * rest procedure of ptlrpc */
524 if (reqmsg_size != 0) {
525 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
527 LASSERT(req->rq_reqmsg);
528 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
530 CWARN("failed to alloc reqbuf: %d\n", rc);
531 req->rq_flvr = old_flvr;
534 OBD_FREE_LARGE(reqmsg, reqmsg_size);
540 * If current context of \a req is dead somehow, e.g. we just switched flavor
541 * thus marked original contexts dead, we'll find a new context for it. if
542 * no switch is needed, \a req will end up with the same context.
544 * \note a request must have a context, to keep other parts of code happy.
545 * In any case of failure during the switching, we must restore the old one.
547 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
549 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
550 struct ptlrpc_cli_ctx *newctx;
556 sptlrpc_cli_ctx_get(oldctx);
557 sptlrpc_req_put_ctx(req, 0);
559 rc = sptlrpc_req_get_ctx(req);
561 LASSERT(!req->rq_cli_ctx);
563 /* restore old ctx */
564 req->rq_cli_ctx = oldctx;
568 newctx = req->rq_cli_ctx;
571 if (unlikely(newctx == oldctx &&
572 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
574 * still get the old dead ctx, usually means system too busy
577 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
578 newctx, newctx->cc_flags);
580 set_current_state(TASK_INTERRUPTIBLE);
581 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
582 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
585 * new ctx not up to date yet
588 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
589 newctx, newctx->cc_flags);
592 * it's possible newctx == oldctx if we're switching
593 * subflavor with the same sec.
595 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
597 /* restore old ctx */
598 sptlrpc_req_put_ctx(req, 0);
599 req->rq_cli_ctx = oldctx;
603 LASSERT(req->rq_cli_ctx == newctx);
606 sptlrpc_cli_ctx_put(oldctx, 1);
609 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
612 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
614 if (cli_ctx_is_refreshed(ctx))
620 int ctx_refresh_timeout(void *data)
622 struct ptlrpc_request *req = data;
625 /* conn_cnt is needed in expire_one_request */
626 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
628 rc = ptlrpc_expire_one_request(req, 1);
629 /* if we started recovery, we should mark this ctx dead; otherwise
630 * in case of lgssd died nobody would retire this ctx, following
631 * connecting will still find the same ctx thus cause deadlock.
632 * there's an assumption that expire time of the request should be
633 * later than the context refresh expire time.
636 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
641 void ctx_refresh_interrupt(void *data)
643 struct ptlrpc_request *req = data;
645 spin_lock(&req->rq_lock);
647 spin_unlock(&req->rq_lock);
651 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
653 spin_lock(&ctx->cc_lock);
654 if (!list_empty(&req->rq_ctx_chain))
655 list_del_init(&req->rq_ctx_chain);
656 spin_unlock(&ctx->cc_lock);
660 * To refresh the context of \req, if it's not up-to-date.
663 * - = 0: wait until success or fatal error occur
664 * - > 0: timeout value (in seconds)
666 * The status of the context could be subject to be changed by other threads
667 * at any time. We allow this race, but once we return with 0, the caller will
668 * suppose it's uptodated and keep using it until the owning rpc is done.
670 * \retval 0 only if the context is uptodated.
671 * \retval -ev error number.
673 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
675 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
676 struct ptlrpc_sec *sec;
677 struct l_wait_info lwi;
683 if (req->rq_ctx_init || req->rq_ctx_fini)
687 * during the process a request's context might change type even
688 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
692 rc = import_sec_validate_get(req->rq_import, &sec);
696 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
697 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
698 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
699 req_off_ctx_list(req, ctx);
700 sptlrpc_req_replace_dead_ctx(req);
701 ctx = req->rq_cli_ctx;
703 sptlrpc_sec_put(sec);
705 if (cli_ctx_is_eternal(ctx))
708 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
709 LASSERT(ctx->cc_ops->refresh);
710 ctx->cc_ops->refresh(ctx);
712 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
714 LASSERT(ctx->cc_ops->validate);
715 if (ctx->cc_ops->validate(ctx) == 0) {
716 req_off_ctx_list(req, ctx);
720 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
721 spin_lock(&req->rq_lock);
723 spin_unlock(&req->rq_lock);
724 req_off_ctx_list(req, ctx);
729 * There's a subtle issue for resending RPCs, suppose following
731 * 1. the request was sent to server.
732 * 2. recovery was kicked start, after finished the request was
734 * 3. resend the request.
735 * 4. old reply from server received, we accept and verify the reply.
736 * this has to be success, otherwise the error will be aware
738 * 5. new reply from server received, dropped by LNet.
740 * Note the xid of old & new request is the same. We can't simply
741 * change xid for the resent request because the server replies on
742 * it for reply reconstruction.
744 * Commonly the original context should be uptodate because we
745 * have an expiry nice time; server will keep its context because
746 * we at least hold a ref of old context which prevent context
747 * from destroying RPC being sent. So server still can accept the
748 * request and finish the RPC. But if that's not the case:
749 * 1. If server side context has been trimmed, a NO_CONTEXT will
750 * be returned, gss_cli_ctx_verify/unseal will switch to new
752 * 2. Current context never be refreshed, then we are fine: we
753 * never really send request with old context before.
755 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
756 unlikely(req->rq_reqmsg) &&
757 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
758 req_off_ctx_list(req, ctx);
762 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
763 req_off_ctx_list(req, ctx);
765 * don't switch ctx if import was deactivated
767 if (req->rq_import->imp_deactive) {
768 spin_lock(&req->rq_lock);
770 spin_unlock(&req->rq_lock);
774 rc = sptlrpc_req_replace_dead_ctx(req);
776 LASSERT(ctx == req->rq_cli_ctx);
777 CERROR("req %p: failed to replace dead ctx %p: %d\n",
779 spin_lock(&req->rq_lock);
781 spin_unlock(&req->rq_lock);
785 ctx = req->rq_cli_ctx;
790 * Now we're sure this context is during upcall, add myself into
793 spin_lock(&ctx->cc_lock);
794 if (list_empty(&req->rq_ctx_chain))
795 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
796 spin_unlock(&ctx->cc_lock);
799 RETURN(-EWOULDBLOCK);
801 /* Clear any flags that may be present from previous sends */
802 LASSERT(req->rq_receiving_reply == 0);
803 spin_lock(&req->rq_lock);
805 req->rq_timedout = 0;
808 spin_unlock(&req->rq_lock);
810 lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
812 ctx_refresh_interrupt, req);
813 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
816 * following cases could lead us here:
817 * - successfully refreshed;
819 * - timedout, and we don't want recover from the failure;
820 * - timedout, and waked up upon recovery finished;
821 * - someone else mark this ctx dead by force;
822 * - someone invalidate the req and call ptlrpc_client_wake_req(),
823 * e.g. ptlrpc_abort_inflight();
825 if (!cli_ctx_is_refreshed(ctx)) {
826 /* timed out or interruptted */
827 req_off_ctx_list(req, ctx);
837 * Initialize flavor settings for \a req, according to \a opcode.
839 * \note this could be called in two situations:
840 * - new request from ptlrpc_pre_req(), with proper @opcode
841 * - old request which changed ctx in the middle, with @opcode == 0
843 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
845 struct ptlrpc_sec *sec;
847 LASSERT(req->rq_import);
848 LASSERT(req->rq_cli_ctx);
849 LASSERT(req->rq_cli_ctx->cc_sec);
850 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
852 /* special security flags accoding to opcode */
856 case MGS_CONFIG_READ:
858 req->rq_bulk_read = 1;
862 req->rq_bulk_write = 1;
865 req->rq_ctx_init = 1;
868 req->rq_ctx_fini = 1;
871 /* init/fini rpc won't be resend, so can't be here */
872 LASSERT(req->rq_ctx_init == 0);
873 LASSERT(req->rq_ctx_fini == 0);
875 /* cleanup flags, which should be recalculated */
876 req->rq_pack_udesc = 0;
877 req->rq_pack_bulk = 0;
881 sec = req->rq_cli_ctx->cc_sec;
883 spin_lock(&sec->ps_lock);
884 req->rq_flvr = sec->ps_flvr;
885 spin_unlock(&sec->ps_lock);
887 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
889 if (unlikely(req->rq_ctx_init))
890 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
891 else if (unlikely(req->rq_ctx_fini))
892 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
894 /* user descriptor flag, null security can't do it anyway */
895 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
896 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
897 req->rq_pack_udesc = 1;
899 /* bulk security flag */
900 if ((req->rq_bulk_read || req->rq_bulk_write) &&
901 sptlrpc_flavor_has_bulk(&req->rq_flvr))
902 req->rq_pack_bulk = 1;
905 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
907 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
910 LASSERT(req->rq_clrbuf);
911 if (req->rq_pool || !req->rq_reqbuf)
914 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
915 req->rq_reqbuf = NULL;
916 req->rq_reqbuf_len = 0;
920 * Given an import \a imp, check whether current user has a valid context
921 * or not. We may create a new context and try to refresh it, and try
922 * repeatedly try in case of non-fatal errors. Return 0 means success.
924 int sptlrpc_import_check_ctx(struct obd_import *imp)
926 struct ptlrpc_sec *sec;
927 struct ptlrpc_cli_ctx *ctx;
928 struct ptlrpc_request *req = NULL;
934 sec = sptlrpc_import_sec_ref(imp);
935 ctx = get_my_ctx(sec);
936 sptlrpc_sec_put(sec);
941 if (cli_ctx_is_eternal(ctx) ||
942 ctx->cc_ops->validate(ctx) == 0) {
943 sptlrpc_cli_ctx_put(ctx, 1);
947 if (cli_ctx_is_error(ctx)) {
948 sptlrpc_cli_ctx_put(ctx, 1);
952 req = ptlrpc_request_cache_alloc(GFP_NOFS);
956 ptlrpc_cli_req_init(req);
957 atomic_set(&req->rq_refcount, 10000);
959 req->rq_import = imp;
960 req->rq_flvr = sec->ps_flvr;
961 req->rq_cli_ctx = ctx;
963 rc = sptlrpc_req_refresh_ctx(req, 0);
964 LASSERT(list_empty(&req->rq_ctx_chain));
965 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
966 ptlrpc_request_cache_free(req);
972 * Used by ptlrpc client, to perform the pre-defined security transformation
973 * upon the request message of \a req. After this function called,
974 * req->rq_reqmsg is still accessible as clear text.
976 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
978 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
983 LASSERT(ctx->cc_sec);
984 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
986 /* we wrap bulk request here because now we can be sure
987 * the context is uptodate.
990 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
995 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
996 case SPTLRPC_SVC_NULL:
997 case SPTLRPC_SVC_AUTH:
998 case SPTLRPC_SVC_INTG:
999 LASSERT(ctx->cc_ops->sign);
1000 rc = ctx->cc_ops->sign(ctx, req);
1002 case SPTLRPC_SVC_PRIV:
1003 LASSERT(ctx->cc_ops->seal);
1004 rc = ctx->cc_ops->seal(ctx, req);
1011 LASSERT(req->rq_reqdata_len);
1012 LASSERT(req->rq_reqdata_len % 8 == 0);
1013 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1019 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1021 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1026 LASSERT(ctx->cc_sec);
1027 LASSERT(req->rq_repbuf);
1028 LASSERT(req->rq_repdata);
1029 LASSERT(req->rq_repmsg == NULL);
1031 req->rq_rep_swab_mask = 0;
1033 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1036 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1040 CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
1044 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1045 CERROR("replied data length %d too small\n",
1046 req->rq_repdata_len);
1050 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1051 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1052 CERROR("reply policy %u doesn't match request policy %u\n",
1053 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1054 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1058 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1059 case SPTLRPC_SVC_NULL:
1060 case SPTLRPC_SVC_AUTH:
1061 case SPTLRPC_SVC_INTG:
1062 LASSERT(ctx->cc_ops->verify);
1063 rc = ctx->cc_ops->verify(ctx, req);
1065 case SPTLRPC_SVC_PRIV:
1066 LASSERT(ctx->cc_ops->unseal);
1067 rc = ctx->cc_ops->unseal(ctx, req);
1072 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1074 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1076 req->rq_rep_swab_mask = 0;
1081 * Used by ptlrpc client, to perform security transformation upon the reply
1082 * message of \a req. After return successfully, req->rq_repmsg points to
1083 * the reply message in clear text.
1085 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1088 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1090 LASSERT(req->rq_repbuf);
1091 LASSERT(req->rq_repdata == NULL);
1092 LASSERT(req->rq_repmsg == NULL);
1093 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1095 if (req->rq_reply_off == 0 &&
1096 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1097 CERROR("real reply with offset 0\n");
1101 if (req->rq_reply_off % 8 != 0) {
1102 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1106 req->rq_repdata = (struct lustre_msg *)
1107 (req->rq_repbuf + req->rq_reply_off);
1108 req->rq_repdata_len = req->rq_nob_received;
1110 return do_cli_unwrap_reply(req);
1114 * Used by ptlrpc client, to perform security transformation upon the early
1115 * reply message of \a req. We expect the rq_reply_off is 0, and
1116 * rq_nob_received is the early reply size.
1118 * Because the receive buffer might be still posted, the reply data might be
1119 * changed at any time, no matter we're holding rq_lock or not. For this reason
1120 * we allocate a separate ptlrpc_request and reply buffer for early reply
1123 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1124 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1125 * \a *req_ret to release it.
1126 * \retval -ev error number, and \a req_ret will not be set.
1128 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1129 struct ptlrpc_request **req_ret)
1131 struct ptlrpc_request *early_req;
1133 int early_bufsz, early_size;
1137 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1138 if (early_req == NULL)
1141 ptlrpc_cli_req_init(early_req);
1143 early_size = req->rq_nob_received;
1144 early_bufsz = size_roundup_power2(early_size);
1145 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1146 if (early_buf == NULL)
1147 GOTO(err_req, rc = -ENOMEM);
1149 /* sanity checkings and copy data out, do it inside spinlock */
1150 spin_lock(&req->rq_lock);
1152 if (req->rq_replied) {
1153 spin_unlock(&req->rq_lock);
1154 GOTO(err_buf, rc = -EALREADY);
1157 LASSERT(req->rq_repbuf);
1158 LASSERT(req->rq_repdata == NULL);
1159 LASSERT(req->rq_repmsg == NULL);
1161 if (req->rq_reply_off != 0) {
1162 CERROR("early reply with offset %u\n", req->rq_reply_off);
1163 spin_unlock(&req->rq_lock);
1164 GOTO(err_buf, rc = -EPROTO);
1167 if (req->rq_nob_received != early_size) {
1168 /* even another early arrived the size should be the same */
1169 CERROR("data size has changed from %u to %u\n",
1170 early_size, req->rq_nob_received);
1171 spin_unlock(&req->rq_lock);
1172 GOTO(err_buf, rc = -EINVAL);
1175 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1176 CERROR("early reply length %d too small\n",
1177 req->rq_nob_received);
1178 spin_unlock(&req->rq_lock);
1179 GOTO(err_buf, rc = -EALREADY);
1182 memcpy(early_buf, req->rq_repbuf, early_size);
1183 spin_unlock(&req->rq_lock);
1185 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1186 early_req->rq_flvr = req->rq_flvr;
1187 early_req->rq_repbuf = early_buf;
1188 early_req->rq_repbuf_len = early_bufsz;
1189 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1190 early_req->rq_repdata_len = early_size;
1191 early_req->rq_early = 1;
1192 early_req->rq_reqmsg = req->rq_reqmsg;
1194 rc = do_cli_unwrap_reply(early_req);
1196 DEBUG_REQ(D_ADAPTTO, early_req,
1197 "error %d unwrap early reply", rc);
1201 LASSERT(early_req->rq_repmsg);
1202 *req_ret = early_req;
1206 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1208 OBD_FREE_LARGE(early_buf, early_bufsz);
1210 ptlrpc_request_cache_free(early_req);
1215 * Used by ptlrpc client, to release a processed early reply \a early_req.
1217 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1219 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1221 LASSERT(early_req->rq_repbuf);
1222 LASSERT(early_req->rq_repdata);
1223 LASSERT(early_req->rq_repmsg);
1225 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1226 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1227 ptlrpc_request_cache_free(early_req);
1230 /**************************************************
1232 **************************************************/
1235 * "fixed" sec (e.g. null) use sec_id < 0
1237 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1239 int sptlrpc_get_next_secid(void)
1241 return atomic_inc_return(&sptlrpc_sec_id);
1243 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1245 /**************************************************
1246 * client side high-level security APIs *
1247 **************************************************/
1249 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1250 int grace, int force)
1252 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1254 LASSERT(policy->sp_cops);
1255 LASSERT(policy->sp_cops->flush_ctx_cache);
1257 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1260 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1262 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1264 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1265 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1266 LASSERT(policy->sp_cops->destroy_sec);
1268 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1270 policy->sp_cops->destroy_sec(sec);
1271 sptlrpc_policy_put(policy);
1274 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1276 sec_cop_destroy_sec(sec);
1278 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1280 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1282 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1284 if (sec->ps_policy->sp_cops->kill_sec) {
1285 sec->ps_policy->sp_cops->kill_sec(sec);
1287 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1291 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1294 atomic_inc(&sec->ps_refcount);
1298 EXPORT_SYMBOL(sptlrpc_sec_get);
1300 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1303 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1305 if (atomic_dec_and_test(&sec->ps_refcount)) {
1306 sptlrpc_gc_del_sec(sec);
1307 sec_cop_destroy_sec(sec);
1311 EXPORT_SYMBOL(sptlrpc_sec_put);
1314 * policy module is responsible for taking refrence of import
1317 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1318 struct ptlrpc_svc_ctx *svc_ctx,
1319 struct sptlrpc_flavor *sf,
1320 enum lustre_sec_part sp)
1322 struct ptlrpc_sec_policy *policy;
1323 struct ptlrpc_sec *sec;
1328 LASSERT(imp->imp_dlm_fake == 1);
1330 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1331 imp->imp_obd->obd_type->typ_name,
1332 imp->imp_obd->obd_name,
1333 sptlrpc_flavor2name(sf, str, sizeof(str)));
1335 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1336 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1338 LASSERT(imp->imp_dlm_fake == 0);
1340 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1341 imp->imp_obd->obd_type->typ_name,
1342 imp->imp_obd->obd_name,
1343 sptlrpc_flavor2name(sf, str, sizeof(str)));
1345 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1347 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1352 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1354 atomic_inc(&sec->ps_refcount);
1358 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1359 sptlrpc_gc_add_sec(sec);
1361 sptlrpc_policy_put(policy);
1367 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1369 struct ptlrpc_sec *sec;
1371 spin_lock(&imp->imp_lock);
1372 sec = sptlrpc_sec_get(imp->imp_sec);
1373 spin_unlock(&imp->imp_lock);
1377 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1379 static void sptlrpc_import_sec_install(struct obd_import *imp,
1380 struct ptlrpc_sec *sec)
1382 struct ptlrpc_sec *old_sec;
1384 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1386 spin_lock(&imp->imp_lock);
1387 old_sec = imp->imp_sec;
1389 spin_unlock(&imp->imp_lock);
1392 sptlrpc_sec_kill(old_sec);
1394 /* balance the ref taken by this import */
1395 sptlrpc_sec_put(old_sec);
1400 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1402 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1406 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1411 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
1412 struct ptlrpc_sec *sec,
1413 struct sptlrpc_flavor *sf)
1415 char str1[32], str2[32];
1417 if (sec->ps_flvr.sf_flags != sf->sf_flags)
1418 CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
1419 sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
1420 str1, sizeof(str1)),
1421 sptlrpc_secflags2str(sf->sf_flags,
1422 str2, sizeof(str2)));
1424 spin_lock(&sec->ps_lock);
1425 flavor_copy(&sec->ps_flvr, sf);
1426 spin_unlock(&sec->ps_lock);
1430 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1431 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1433 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1434 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1436 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1437 struct ptlrpc_svc_ctx *svc_ctx,
1438 struct sptlrpc_flavor *flvr)
1440 struct ptlrpc_connection *conn;
1441 struct sptlrpc_flavor sf;
1442 struct ptlrpc_sec *sec, *newsec;
1443 enum lustre_sec_part sp;
1453 conn = imp->imp_connection;
1455 if (svc_ctx == NULL) {
1456 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1458 * normal import, determine flavor from rule set, except
1459 * for mgc the flavor is predetermined.
1461 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1462 sf = cliobd->cl_flvr_mgc;
1464 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1466 &cliobd->cl_target_uuid,
1469 sp = imp->imp_obd->u.cli.cl_sp_me;
1471 /* reverse import, determine flavor from incoming reqeust */
1474 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1475 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1476 PTLRPC_SEC_FL_ROOTONLY;
1478 sp = sptlrpc_target_sec_part(imp->imp_obd);
1481 sec = sptlrpc_import_sec_ref(imp);
1485 if (flavor_equal(&sf, &sec->ps_flvr))
1488 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1489 imp->imp_obd->obd_name,
1490 obd_uuid2str(&conn->c_remote_uuid),
1491 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1492 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1494 if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
1495 SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
1496 SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
1497 SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
1498 sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
1501 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1502 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1503 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1504 imp->imp_obd->obd_name,
1505 obd_uuid2str(&conn->c_remote_uuid),
1506 LNET_NIDNET(conn->c_self),
1507 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1510 mutex_lock(&imp->imp_sec_mutex);
1512 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1514 sptlrpc_import_sec_install(imp, newsec);
1516 CERROR("import %s->%s: failed to create new sec\n",
1517 imp->imp_obd->obd_name,
1518 obd_uuid2str(&conn->c_remote_uuid));
1522 mutex_unlock(&imp->imp_sec_mutex);
1524 sptlrpc_sec_put(sec);
1528 void sptlrpc_import_sec_put(struct obd_import *imp)
1531 sptlrpc_sec_kill(imp->imp_sec);
1533 sptlrpc_sec_put(imp->imp_sec);
1534 imp->imp_sec = NULL;
1538 static void import_flush_ctx_common(struct obd_import *imp,
1539 uid_t uid, int grace, int force)
1541 struct ptlrpc_sec *sec;
1546 sec = sptlrpc_import_sec_ref(imp);
1550 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1551 sptlrpc_sec_put(sec);
1554 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1556 /* it's important to use grace mode, see explain in
1557 * sptlrpc_req_refresh_ctx() */
1558 import_flush_ctx_common(imp, 0, 1, 1);
1561 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1563 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1566 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1568 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1570 import_flush_ctx_common(imp, -1, 1, 1);
1572 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1575 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1576 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1578 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1580 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1581 struct ptlrpc_sec_policy *policy;
1585 LASSERT(ctx->cc_sec);
1586 LASSERT(ctx->cc_sec->ps_policy);
1587 LASSERT(req->rq_reqmsg == NULL);
1588 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1590 policy = ctx->cc_sec->ps_policy;
1591 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1593 LASSERT(req->rq_reqmsg);
1594 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1596 /* zeroing preallocated buffer */
1598 memset(req->rq_reqmsg, 0, msgsize);
1605 * Used by ptlrpc client to free request buffer of \a req. After this
1606 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1608 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1610 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1611 struct ptlrpc_sec_policy *policy;
1614 LASSERT(ctx->cc_sec);
1615 LASSERT(ctx->cc_sec->ps_policy);
1616 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1618 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1621 policy = ctx->cc_sec->ps_policy;
1622 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1623 req->rq_reqmsg = NULL;
1627 * NOTE caller must guarantee the buffer size is enough for the enlargement
1629 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1630 int segment, int newsize)
1633 int oldsize, oldmsg_size, movesize;
1635 LASSERT(segment < msg->lm_bufcount);
1636 LASSERT(msg->lm_buflens[segment] <= newsize);
1638 if (msg->lm_buflens[segment] == newsize)
1641 /* nothing to do if we are enlarging the last segment */
1642 if (segment == msg->lm_bufcount - 1) {
1643 msg->lm_buflens[segment] = newsize;
1647 oldsize = msg->lm_buflens[segment];
1649 src = lustre_msg_buf(msg, segment + 1, 0);
1650 msg->lm_buflens[segment] = newsize;
1651 dst = lustre_msg_buf(msg, segment + 1, 0);
1652 msg->lm_buflens[segment] = oldsize;
1654 /* move from segment + 1 to end segment */
1655 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1656 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1657 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1658 LASSERT(movesize >= 0);
1661 memmove(dst, src, movesize);
1663 /* note we don't clear the ares where old data live, not secret */
1665 /* finally set new segment size */
1666 msg->lm_buflens[segment] = newsize;
1668 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1671 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1672 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1673 * preserved after the enlargement. this must be called after original request
1674 * buffer being allocated.
1676 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1677 * so caller should refresh its local pointers if needed.
1679 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1680 int segment, int newsize)
1682 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1683 struct ptlrpc_sec_cops *cops;
1684 struct lustre_msg *msg = req->rq_reqmsg;
1688 LASSERT(msg->lm_bufcount > segment);
1689 LASSERT(msg->lm_buflens[segment] <= newsize);
1691 if (msg->lm_buflens[segment] == newsize)
1694 cops = ctx->cc_sec->ps_policy->sp_cops;
1695 LASSERT(cops->enlarge_reqbuf);
1696 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1698 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1701 * Used by ptlrpc client to allocate reply buffer of \a req.
1703 * \note After this, req->rq_repmsg is still not accessible.
1705 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1707 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1708 struct ptlrpc_sec_policy *policy;
1712 LASSERT(ctx->cc_sec);
1713 LASSERT(ctx->cc_sec->ps_policy);
1718 policy = ctx->cc_sec->ps_policy;
1719 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1723 * Used by ptlrpc client to free reply buffer of \a req. After this
1724 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1726 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1728 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1729 struct ptlrpc_sec_policy *policy;
1733 LASSERT(ctx->cc_sec);
1734 LASSERT(ctx->cc_sec->ps_policy);
1735 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1737 if (req->rq_repbuf == NULL)
1739 LASSERT(req->rq_repbuf_len);
1741 policy = ctx->cc_sec->ps_policy;
1742 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1743 req->rq_repmsg = NULL;
1747 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1748 struct ptlrpc_cli_ctx *ctx)
1750 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1752 if (!policy->sp_cops->install_rctx)
1754 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1757 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1758 struct ptlrpc_svc_ctx *ctx)
1760 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1762 if (!policy->sp_sops->install_rctx)
1764 return policy->sp_sops->install_rctx(imp, ctx);
1767 /****************************************
1768 * server side security *
1769 ****************************************/
1771 static int flavor_allowed(struct sptlrpc_flavor *exp,
1772 struct ptlrpc_request *req)
1774 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1776 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1779 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1780 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1781 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1782 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1788 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1791 * Given an export \a exp, check whether the flavor of incoming \a req
1792 * is allowed by the export \a exp. Main logic is about taking care of
1793 * changing configurations. Return 0 means success.
1795 int sptlrpc_target_export_check(struct obd_export *exp,
1796 struct ptlrpc_request *req)
1798 struct sptlrpc_flavor flavor;
1803 /* client side export has no imp_reverse, skip
1804 * FIXME maybe we should check flavor this as well??? */
1805 if (exp->exp_imp_reverse == NULL)
1808 /* don't care about ctx fini rpc */
1809 if (req->rq_ctx_fini)
1812 spin_lock(&exp->exp_lock);
1814 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1815 * the first req with the new flavor, then treat it as current flavor,
1816 * adapt reverse sec according to it.
1817 * note the first rpc with new flavor might not be with root ctx, in
1818 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1819 if (unlikely(exp->exp_flvr_changed) &&
1820 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1821 /* make the new flavor as "current", and old ones as
1822 * about-to-expire */
1823 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1824 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1825 flavor = exp->exp_flvr_old[1];
1826 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1827 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1828 exp->exp_flvr_old[0] = exp->exp_flvr;
1829 exp->exp_flvr_expire[0] = cfs_time_current_sec() +
1830 EXP_FLVR_UPDATE_EXPIRE;
1831 exp->exp_flvr = flavor;
1833 /* flavor change finished */
1834 exp->exp_flvr_changed = 0;
1835 LASSERT(exp->exp_flvr_adapt == 1);
1837 /* if it's gss, we only interested in root ctx init */
1838 if (req->rq_auth_gss &&
1839 !(req->rq_ctx_init &&
1840 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1841 req->rq_auth_usr_ost))) {
1842 spin_unlock(&exp->exp_lock);
1843 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1844 req->rq_auth_gss, req->rq_ctx_init,
1845 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1846 req->rq_auth_usr_ost);
1850 exp->exp_flvr_adapt = 0;
1851 spin_unlock(&exp->exp_lock);
1853 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1854 req->rq_svc_ctx, &flavor);
1857 /* if it equals to the current flavor, we accept it, but need to
1858 * dealing with reverse sec/ctx */
1859 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1860 /* most cases should return here, we only interested in
1861 * gss root ctx init */
1862 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1863 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1864 !req->rq_auth_usr_ost)) {
1865 spin_unlock(&exp->exp_lock);
1869 /* if flavor just changed, we should not proceed, just leave
1870 * it and current flavor will be discovered and replaced
1871 * shortly, and let _this_ rpc pass through */
1872 if (exp->exp_flvr_changed) {
1873 LASSERT(exp->exp_flvr_adapt);
1874 spin_unlock(&exp->exp_lock);
1878 if (exp->exp_flvr_adapt) {
1879 exp->exp_flvr_adapt = 0;
1880 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1881 exp, exp->exp_flvr.sf_rpc,
1882 exp->exp_flvr_old[0].sf_rpc,
1883 exp->exp_flvr_old[1].sf_rpc);
1884 flavor = exp->exp_flvr;
1885 spin_unlock(&exp->exp_lock);
1887 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1891 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1892 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1893 exp->exp_flvr_old[0].sf_rpc,
1894 exp->exp_flvr_old[1].sf_rpc);
1895 spin_unlock(&exp->exp_lock);
1897 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1902 if (exp->exp_flvr_expire[0]) {
1903 if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
1904 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1905 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1906 "middle 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[0] -
1911 cfs_time_current_sec());
1912 spin_unlock(&exp->exp_lock);
1916 CDEBUG(D_SEC, "mark middle expired\n");
1917 exp->exp_flvr_expire[0] = 0;
1919 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1920 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);
1925 /* now it doesn't match the current flavor, the only chance we can
1926 * accept it is match the old flavors which is not expired. */
1927 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1928 if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
1929 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1930 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
1931 "oldest one ("CFS_DURATION_T")\n", exp,
1932 exp->exp_flvr.sf_rpc,
1933 exp->exp_flvr_old[0].sf_rpc,
1934 exp->exp_flvr_old[1].sf_rpc,
1935 exp->exp_flvr_expire[1] -
1936 cfs_time_current_sec());
1937 spin_unlock(&exp->exp_lock);
1941 CDEBUG(D_SEC, "mark oldest expired\n");
1942 exp->exp_flvr_expire[1] = 0;
1944 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1945 exp, exp->exp_flvr.sf_rpc,
1946 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1947 req->rq_flvr.sf_rpc);
1949 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1950 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1951 exp->exp_flvr_old[1].sf_rpc);
1954 spin_unlock(&exp->exp_lock);
1956 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
1957 "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
1958 exp, exp->exp_obd->obd_name,
1959 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1960 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1961 req->rq_flvr.sf_rpc,
1962 exp->exp_flvr.sf_rpc,
1963 exp->exp_flvr_old[0].sf_rpc,
1964 exp->exp_flvr_expire[0] ?
1965 (unsigned long) (exp->exp_flvr_expire[0] -
1966 cfs_time_current_sec()) : 0,
1967 exp->exp_flvr_old[1].sf_rpc,
1968 exp->exp_flvr_expire[1] ?
1969 (unsigned long) (exp->exp_flvr_expire[1] -
1970 cfs_time_current_sec()) : 0);
1973 EXPORT_SYMBOL(sptlrpc_target_export_check);
1975 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1976 struct sptlrpc_rule_set *rset)
1978 struct obd_export *exp;
1979 struct sptlrpc_flavor new_flvr;
1983 spin_lock(&obd->obd_dev_lock);
1985 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1986 if (exp->exp_connection == NULL)
1989 /* note if this export had just been updated flavor
1990 * (exp_flvr_changed == 1), this will override the
1992 spin_lock(&exp->exp_lock);
1993 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1994 exp->exp_connection->c_peer.nid,
1996 if (exp->exp_flvr_changed ||
1997 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1998 exp->exp_flvr_old[1] = new_flvr;
1999 exp->exp_flvr_expire[1] = 0;
2000 exp->exp_flvr_changed = 1;
2001 exp->exp_flvr_adapt = 1;
2003 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2004 exp, sptlrpc_part2name(exp->exp_sp_peer),
2005 exp->exp_flvr.sf_rpc,
2006 exp->exp_flvr_old[1].sf_rpc);
2008 spin_unlock(&exp->exp_lock);
2011 spin_unlock(&obd->obd_dev_lock);
2013 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2015 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2017 /* peer's claim is unreliable unless gss is being used */
2018 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2021 switch (req->rq_sp_from) {
2023 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2024 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2025 svc_rc = SECSVC_DROP;
2029 if (!req->rq_auth_usr_mdt) {
2030 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2031 svc_rc = SECSVC_DROP;
2035 if (!req->rq_auth_usr_ost) {
2036 DEBUG_REQ(D_ERROR, req, "faked source OST");
2037 svc_rc = SECSVC_DROP;
2042 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2043 !req->rq_auth_usr_ost) {
2044 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2045 svc_rc = SECSVC_DROP;
2050 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2051 svc_rc = SECSVC_DROP;
2058 * Used by ptlrpc server, to perform transformation upon request message of
2059 * incoming \a req. This must be the first thing to do with an incoming
2060 * request in ptlrpc layer.
2062 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2063 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2064 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2065 * reply message has been prepared.
2066 * \retval SECSVC_DROP failed, this request should be dropped.
2068 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2070 struct ptlrpc_sec_policy *policy;
2071 struct lustre_msg *msg = req->rq_reqbuf;
2076 LASSERT(req->rq_reqmsg == NULL);
2077 LASSERT(req->rq_repmsg == NULL);
2078 LASSERT(req->rq_svc_ctx == NULL);
2080 req->rq_req_swab_mask = 0;
2082 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2085 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2089 CERROR("error unpacking request from %s x"LPU64"\n",
2090 libcfs_id2str(req->rq_peer), req->rq_xid);
2091 RETURN(SECSVC_DROP);
2094 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2095 req->rq_sp_from = LUSTRE_SP_ANY;
2096 req->rq_auth_uid = -1; /* set to INVALID_UID */
2097 req->rq_auth_mapped_uid = -1;
2099 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2101 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2102 RETURN(SECSVC_DROP);
2105 LASSERT(policy->sp_sops->accept);
2106 rc = policy->sp_sops->accept(req);
2107 sptlrpc_policy_put(policy);
2108 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2109 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2112 * if it's not null flavor (which means embedded packing msg),
2113 * reset the swab mask for the comming inner msg unpacking.
2115 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2116 req->rq_req_swab_mask = 0;
2118 /* sanity check for the request source */
2119 rc = sptlrpc_svc_check_from(req, rc);
2124 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2125 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2126 * a buffer of \a msglen size.
2128 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2130 struct ptlrpc_sec_policy *policy;
2131 struct ptlrpc_reply_state *rs;
2135 LASSERT(req->rq_svc_ctx);
2136 LASSERT(req->rq_svc_ctx->sc_policy);
2138 policy = req->rq_svc_ctx->sc_policy;
2139 LASSERT(policy->sp_sops->alloc_rs);
2141 rc = policy->sp_sops->alloc_rs(req, msglen);
2142 if (unlikely(rc == -ENOMEM)) {
2143 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2144 if (svcpt->scp_service->srv_max_reply_size <
2145 msglen + sizeof(struct ptlrpc_reply_state)) {
2146 /* Just return failure if the size is too big */
2147 CERROR("size of message is too big (%zd), %d allowed\n",
2148 msglen + sizeof(struct ptlrpc_reply_state),
2149 svcpt->scp_service->srv_max_reply_size);
2153 /* failed alloc, try emergency pool */
2154 rs = lustre_get_emerg_rs(svcpt);
2158 req->rq_reply_state = rs;
2159 rc = policy->sp_sops->alloc_rs(req, msglen);
2161 lustre_put_emerg_rs(rs);
2162 req->rq_reply_state = NULL;
2167 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2173 * Used by ptlrpc server, to perform transformation upon reply message.
2175 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2176 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2178 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2180 struct ptlrpc_sec_policy *policy;
2184 LASSERT(req->rq_svc_ctx);
2185 LASSERT(req->rq_svc_ctx->sc_policy);
2187 policy = req->rq_svc_ctx->sc_policy;
2188 LASSERT(policy->sp_sops->authorize);
2190 rc = policy->sp_sops->authorize(req);
2191 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2197 * Used by ptlrpc server, to free reply_state.
2199 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2201 struct ptlrpc_sec_policy *policy;
2202 unsigned int prealloc;
2205 LASSERT(rs->rs_svc_ctx);
2206 LASSERT(rs->rs_svc_ctx->sc_policy);
2208 policy = rs->rs_svc_ctx->sc_policy;
2209 LASSERT(policy->sp_sops->free_rs);
2211 prealloc = rs->rs_prealloc;
2212 policy->sp_sops->free_rs(rs);
2215 lustre_put_emerg_rs(rs);
2219 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2221 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2224 atomic_inc(&ctx->sc_refcount);
2227 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2229 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2234 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2235 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2236 if (ctx->sc_policy->sp_sops->free_ctx)
2237 ctx->sc_policy->sp_sops->free_ctx(ctx);
2239 req->rq_svc_ctx = NULL;
2242 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2244 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2249 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2250 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2251 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2253 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2255 /****************************************
2257 ****************************************/
2260 * Perform transformation upon bulk data pointed by \a desc. This is called
2261 * before transforming the request message.
2263 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2264 struct ptlrpc_bulk_desc *desc)
2266 struct ptlrpc_cli_ctx *ctx;
2268 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2270 if (!req->rq_pack_bulk)
2273 ctx = req->rq_cli_ctx;
2274 if (ctx->cc_ops->wrap_bulk)
2275 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2278 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2281 * This is called after unwrap the reply message.
2282 * return nob of actual plain text size received, or error code.
2284 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2285 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)
2294 return desc->bd_nob_transferred;
2296 ctx = req->rq_cli_ctx;
2297 if (ctx->cc_ops->unwrap_bulk) {
2298 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2302 return desc->bd_nob_transferred;
2304 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2307 * This is called after unwrap the reply message.
2308 * return 0 for success or error code.
2310 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2311 struct ptlrpc_bulk_desc *desc)
2313 struct ptlrpc_cli_ctx *ctx;
2316 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2318 if (!req->rq_pack_bulk)
2321 ctx = req->rq_cli_ctx;
2322 if (ctx->cc_ops->unwrap_bulk) {
2323 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2329 * if everything is going right, nob should equals to nob_transferred.
2330 * in case of privacy mode, nob_transferred needs to be adjusted.
2332 if (desc->bd_nob != desc->bd_nob_transferred) {
2333 CERROR("nob %d doesn't match transferred nob %d\n",
2334 desc->bd_nob, desc->bd_nob_transferred);
2340 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2342 #ifdef HAVE_SERVER_SUPPORT
2344 * Performe transformation upon outgoing bulk read.
2346 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2347 struct ptlrpc_bulk_desc *desc)
2349 struct ptlrpc_svc_ctx *ctx;
2351 LASSERT(req->rq_bulk_read);
2353 if (!req->rq_pack_bulk)
2356 ctx = req->rq_svc_ctx;
2357 if (ctx->sc_policy->sp_sops->wrap_bulk)
2358 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2362 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2365 * Performe transformation upon incoming bulk write.
2367 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2368 struct ptlrpc_bulk_desc *desc)
2370 struct ptlrpc_svc_ctx *ctx;
2373 LASSERT(req->rq_bulk_write);
2376 * if it's in privacy mode, transferred should >= expected; otherwise
2377 * transferred should == expected.
2379 if (desc->bd_nob_transferred < desc->bd_nob ||
2380 (desc->bd_nob_transferred > desc->bd_nob &&
2381 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2382 SPTLRPC_BULK_SVC_PRIV)) {
2383 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2384 desc->bd_nob_transferred, desc->bd_nob);
2388 if (!req->rq_pack_bulk)
2391 ctx = req->rq_svc_ctx;
2392 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2393 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2395 CERROR("error unwrap bulk: %d\n", rc);
2398 /* return 0 to allow reply be sent */
2401 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2404 * Prepare buffers for incoming bulk write.
2406 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2407 struct ptlrpc_bulk_desc *desc)
2409 struct ptlrpc_svc_ctx *ctx;
2411 LASSERT(req->rq_bulk_write);
2413 if (!req->rq_pack_bulk)
2416 ctx = req->rq_svc_ctx;
2417 if (ctx->sc_policy->sp_sops->prep_bulk)
2418 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2422 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2424 #endif /* HAVE_SERVER_SUPPORT */
2426 /****************************************
2427 * user descriptor helpers *
2428 ****************************************/
2430 int sptlrpc_current_user_desc_size(void)
2434 ngroups = current_ngroups;
2436 if (ngroups > LUSTRE_MAX_GROUPS)
2437 ngroups = LUSTRE_MAX_GROUPS;
2438 return sptlrpc_user_desc_size(ngroups);
2440 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2442 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2444 struct ptlrpc_user_desc *pud;
2446 pud = lustre_msg_buf(msg, offset, 0);
2448 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2449 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2450 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2451 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2452 pud->pud_cap = cfs_curproc_cap_pack();
2453 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2456 if (pud->pud_ngroups > current_ngroups)
2457 pud->pud_ngroups = current_ngroups;
2458 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2459 pud->pud_ngroups * sizeof(__u32));
2460 task_unlock(current);
2464 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2466 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2468 struct ptlrpc_user_desc *pud;
2471 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2476 __swab32s(&pud->pud_uid);
2477 __swab32s(&pud->pud_gid);
2478 __swab32s(&pud->pud_fsuid);
2479 __swab32s(&pud->pud_fsgid);
2480 __swab32s(&pud->pud_cap);
2481 __swab32s(&pud->pud_ngroups);
2484 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2485 CERROR("%u groups is too large\n", pud->pud_ngroups);
2489 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2490 msg->lm_buflens[offset]) {
2491 CERROR("%u groups are claimed but bufsize only %u\n",
2492 pud->pud_ngroups, msg->lm_buflens[offset]);
2497 for (i = 0; i < pud->pud_ngroups; i++)
2498 __swab32s(&pud->pud_groups[i]);
2503 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2505 /****************************************
2507 ****************************************/
2509 const char * sec2target_str(struct ptlrpc_sec *sec)
2511 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2513 if (sec_is_reverse(sec))
2515 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2517 EXPORT_SYMBOL(sec2target_str);
2520 * return true if the bulk data is protected
2522 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2524 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2525 case SPTLRPC_BULK_SVC_INTG:
2526 case SPTLRPC_BULK_SVC_PRIV:
2532 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2534 /****************************************
2535 * crypto API helper/alloc blkciper *
2536 ****************************************/
2538 /****************************************
2539 * initialize/finalize *
2540 ****************************************/
2542 int sptlrpc_init(void)
2546 rwlock_init(&policy_lock);
2548 rc = sptlrpc_gc_init();
2552 rc = sptlrpc_conf_init();
2556 rc = sptlrpc_enc_pool_init();
2560 rc = sptlrpc_null_init();
2564 rc = sptlrpc_plain_init();
2568 rc = sptlrpc_lproc_init();
2575 sptlrpc_plain_fini();
2577 sptlrpc_null_fini();
2579 sptlrpc_enc_pool_fini();
2581 sptlrpc_conf_fini();
2588 void sptlrpc_fini(void)
2590 sptlrpc_lproc_fini();
2591 sptlrpc_plain_fini();
2592 sptlrpc_null_fini();
2593 sptlrpc_enc_pool_fini();
2594 sptlrpc_conf_fini();