4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
33 * Author: Eric Mei <ericm@clusterfs.com>
36 #define DEBUG_SUBSYSTEM S_SEC
38 #include <linux/user_namespace.h>
39 #include <linux/uidgid.h>
40 #include <linux/crypto.h>
41 #include <linux/key.h>
43 #include <libcfs/libcfs.h>
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <lustre_net.h>
48 #include <lustre_import.h>
49 #include <lustre_dlm.h>
50 #include <lustre_sec.h>
52 #include "ptlrpc_internal.h"
54 static int send_sepol;
55 module_param(send_sepol, int, 0644);
56 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
62 static rwlock_t policy_lock;
63 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
67 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
69 __u16 number = policy->sp_policy;
71 LASSERT(policy->sp_name);
72 LASSERT(policy->sp_cops);
73 LASSERT(policy->sp_sops);
75 if (number >= SPTLRPC_POLICY_MAX)
78 write_lock(&policy_lock);
79 if (unlikely(policies[number])) {
80 write_unlock(&policy_lock);
83 policies[number] = policy;
84 write_unlock(&policy_lock);
86 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
89 EXPORT_SYMBOL(sptlrpc_register_policy);
91 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
93 __u16 number = policy->sp_policy;
95 LASSERT(number < SPTLRPC_POLICY_MAX);
97 write_lock(&policy_lock);
98 if (unlikely(policies[number] == NULL)) {
99 write_unlock(&policy_lock);
100 CERROR("%s: already unregistered\n", policy->sp_name);
104 LASSERT(policies[number] == policy);
105 policies[number] = NULL;
106 write_unlock(&policy_lock);
108 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
111 EXPORT_SYMBOL(sptlrpc_unregister_policy);
114 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
116 static DEFINE_MUTEX(load_mutex);
117 static atomic_t loaded = ATOMIC_INIT(0);
118 struct ptlrpc_sec_policy *policy;
119 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
122 if (number >= SPTLRPC_POLICY_MAX)
126 read_lock(&policy_lock);
127 policy = policies[number];
128 if (policy && !try_module_get(policy->sp_owner))
131 flag = atomic_read(&loaded);
132 read_unlock(&policy_lock);
134 if (policy != NULL || flag != 0 ||
135 number != SPTLRPC_POLICY_GSS)
138 /* try to load gss module, once */
139 mutex_lock(&load_mutex);
140 if (atomic_read(&loaded) == 0) {
141 if (request_module("ptlrpc_gss") == 0)
143 "module ptlrpc_gss loaded on demand\n");
145 CERROR("Unable to load module ptlrpc_gss\n");
147 atomic_set(&loaded, 1);
149 mutex_unlock(&load_mutex);
155 __u32 sptlrpc_name2flavor_base(const char *name)
157 if (!strcmp(name, "null"))
158 return SPTLRPC_FLVR_NULL;
159 if (!strcmp(name, "plain"))
160 return SPTLRPC_FLVR_PLAIN;
161 if (!strcmp(name, "gssnull"))
162 return SPTLRPC_FLVR_GSSNULL;
163 if (!strcmp(name, "krb5n"))
164 return SPTLRPC_FLVR_KRB5N;
165 if (!strcmp(name, "krb5a"))
166 return SPTLRPC_FLVR_KRB5A;
167 if (!strcmp(name, "krb5i"))
168 return SPTLRPC_FLVR_KRB5I;
169 if (!strcmp(name, "krb5p"))
170 return SPTLRPC_FLVR_KRB5P;
171 if (!strcmp(name, "skn"))
172 return SPTLRPC_FLVR_SKN;
173 if (!strcmp(name, "ska"))
174 return SPTLRPC_FLVR_SKA;
175 if (!strcmp(name, "ski"))
176 return SPTLRPC_FLVR_SKI;
177 if (!strcmp(name, "skpi"))
178 return SPTLRPC_FLVR_SKPI;
180 return SPTLRPC_FLVR_INVALID;
182 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
184 const char *sptlrpc_flavor2name_base(__u32 flvr)
186 __u32 base = SPTLRPC_FLVR_BASE(flvr);
188 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
190 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
192 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
194 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
196 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
198 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
200 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
202 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
204 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
206 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
208 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
211 CERROR("invalid wire flavor 0x%x\n", flvr);
214 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
216 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
217 char *buf, int bufsize)
219 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
220 snprintf(buf, bufsize, "hash:%s",
221 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
223 snprintf(buf, bufsize, "%s",
224 sptlrpc_flavor2name_base(sf->sf_rpc));
226 buf[bufsize - 1] = '\0';
229 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
231 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
233 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
236 * currently we don't support customized bulk specification for
237 * flavors other than plain
239 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
243 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
244 strncat(buf, bspec, bufsize);
247 buf[bufsize - 1] = '\0';
250 EXPORT_SYMBOL(sptlrpc_flavor2name);
252 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
256 if (flags & PTLRPC_SEC_FL_REVERSE)
257 strlcat(buf, "reverse,", bufsize);
258 if (flags & PTLRPC_SEC_FL_ROOTONLY)
259 strlcat(buf, "rootonly,", bufsize);
260 if (flags & PTLRPC_SEC_FL_UDESC)
261 strlcat(buf, "udesc,", bufsize);
262 if (flags & PTLRPC_SEC_FL_BULK)
263 strlcat(buf, "bulk,", bufsize);
265 strlcat(buf, "-,", bufsize);
269 EXPORT_SYMBOL(sptlrpc_secflags2str);
272 * client context APIs
276 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
278 struct vfs_cred vcred;
279 int create = 1, remove_dead = 1;
282 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
284 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
285 PTLRPC_SEC_FL_ROOTONLY)) {
288 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
293 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
294 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
297 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
301 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
303 atomic_inc(&ctx->cc_refcount);
306 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
308 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
310 struct ptlrpc_sec *sec = ctx->cc_sec;
313 LASSERT(atomic_read(&(ctx)->cc_refcount) > 0);
315 if (!atomic_dec_and_test(&ctx->cc_refcount))
318 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
320 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
323 * Expire the client context immediately.
325 * \pre Caller must hold at least 1 reference on the \a ctx.
327 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
329 LASSERT(ctx->cc_ops->die);
330 ctx->cc_ops->die(ctx, 0);
332 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
335 * To wake up the threads who are waiting for this client context. Called
336 * after some status change happened on \a ctx.
338 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
340 struct ptlrpc_request *req, *next;
342 spin_lock(&ctx->cc_lock);
343 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
345 list_del_init(&req->rq_ctx_chain);
346 ptlrpc_client_wake_req(req);
348 spin_unlock(&ctx->cc_lock);
350 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
352 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
354 LASSERT(ctx->cc_ops);
356 if (ctx->cc_ops->display == NULL)
359 return ctx->cc_ops->display(ctx, buf, bufsize);
362 static int import_sec_check_expire(struct obd_import *imp)
366 write_lock(&imp->imp_sec_lock);
367 if (imp->imp_sec_expire &&
368 imp->imp_sec_expire < ktime_get_real_seconds()) {
370 imp->imp_sec_expire = 0;
372 write_unlock(&imp->imp_sec_lock);
377 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
378 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
382 * Get and validate the client side ptlrpc security facilities from
383 * \a imp. There is a race condition on client reconnect when the import is
384 * being destroyed while there are outstanding client bound requests. In
385 * this case do not output any error messages if import secuity is not
388 * \param[in] imp obd import associated with client
389 * \param[out] sec client side ptlrpc security
391 * \retval 0 if security retrieved successfully
392 * \retval -ve errno if there was a problem
394 static int import_sec_validate_get(struct obd_import *imp,
395 struct ptlrpc_sec **sec)
399 if (unlikely(imp->imp_sec_expire)) {
400 rc = import_sec_check_expire(imp);
405 *sec = sptlrpc_import_sec_ref(imp);
407 /* Only output an error when the import is still active */
408 if (!test_bit(WORK_STRUCT_PENDING_BIT,
409 work_data_bits(&imp->imp_zombie_work)))
410 CERROR("import %p (%s) with no sec\n",
411 imp, ptlrpc_import_state_name(imp->imp_state));
415 if (unlikely((*sec)->ps_dying)) {
416 CERROR("attempt to use dying sec %p\n", sec);
417 sptlrpc_sec_put(*sec);
425 * Given a \a req, find or allocate an appropriate context for it.
426 * \pre req->rq_cli_ctx == NULL.
428 * \retval 0 succeed, and req->rq_cli_ctx is set.
429 * \retval -ev error number, and req->rq_cli_ctx == NULL.
431 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
433 struct obd_import *imp = req->rq_import;
434 struct ptlrpc_sec *sec;
439 LASSERT(!req->rq_cli_ctx);
442 rc = import_sec_validate_get(imp, &sec);
446 req->rq_cli_ctx = get_my_ctx(sec);
448 sptlrpc_sec_put(sec);
450 if (!req->rq_cli_ctx) {
451 CERROR("req %p: fail to get context\n", req);
452 RETURN(-ECONNREFUSED);
459 * Drop the context for \a req.
460 * \pre req->rq_cli_ctx != NULL.
461 * \post req->rq_cli_ctx == NULL.
463 * If \a sync == 0, this function should return quickly without sleep;
464 * otherwise it might trigger and wait for the whole process of sending
465 * an context-destroying rpc to server.
467 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
472 LASSERT(req->rq_cli_ctx);
475 * request might be asked to release earlier while still
476 * in the context waiting list.
478 if (!list_empty(&req->rq_ctx_chain)) {
479 spin_lock(&req->rq_cli_ctx->cc_lock);
480 list_del_init(&req->rq_ctx_chain);
481 spin_unlock(&req->rq_cli_ctx->cc_lock);
484 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
485 req->rq_cli_ctx = NULL;
490 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
491 struct ptlrpc_cli_ctx *oldctx,
492 struct ptlrpc_cli_ctx *newctx)
494 struct sptlrpc_flavor old_flvr;
495 char *reqmsg = NULL; /* to workaround old gcc */
500 "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
501 req, oldctx, oldctx->cc_vcred.vc_uid,
502 sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
503 sec2target_str(newctx->cc_sec), oldctx->cc_sec,
504 oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
505 newctx->cc_sec->ps_policy->sp_name);
508 old_flvr = req->rq_flvr;
510 /* save request message */
511 reqmsg_size = req->rq_reqlen;
512 if (reqmsg_size != 0) {
513 LASSERT(req->rq_reqmsg);
514 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
517 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
520 /* release old req/rep buf */
521 req->rq_cli_ctx = oldctx;
522 sptlrpc_cli_free_reqbuf(req);
523 sptlrpc_cli_free_repbuf(req);
524 req->rq_cli_ctx = newctx;
526 /* recalculate the flavor */
527 sptlrpc_req_set_flavor(req, 0);
530 * alloc new request buffer
531 * we don't need to alloc reply buffer here, leave it to the
532 * rest procedure of ptlrpc
534 if (reqmsg_size != 0) {
535 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
537 LASSERT(req->rq_reqmsg);
538 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
540 CWARN("failed to alloc reqbuf: %d\n", rc);
541 req->rq_flvr = old_flvr;
544 OBD_FREE_LARGE(reqmsg, reqmsg_size);
550 * If current context of \a req is dead somehow, e.g. we just switched flavor
551 * thus marked original contexts dead, we'll find a new context for it. if
552 * no switch is needed, \a req will end up with the same context.
554 * \note a request must have a context, to keep other parts of code happy.
555 * In any case of failure during the switching, we must restore the old one.
557 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
559 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
560 struct ptlrpc_cli_ctx *newctx;
567 sptlrpc_cli_ctx_get(oldctx);
568 sptlrpc_req_put_ctx(req, 0);
570 rc = sptlrpc_req_get_ctx(req);
572 LASSERT(!req->rq_cli_ctx);
574 /* restore old ctx */
575 req->rq_cli_ctx = oldctx;
579 newctx = req->rq_cli_ctx;
582 if (unlikely(newctx == oldctx &&
583 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
585 * still get the old dead ctx, usually means system too busy
588 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
589 newctx, newctx->cc_flags);
591 schedule_timeout_interruptible(cfs_time_seconds(1));
592 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
595 * new ctx not up to date yet
598 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
599 newctx, newctx->cc_flags);
602 * it's possible newctx == oldctx if we're switching
603 * subflavor with the same sec.
605 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
607 /* restore old ctx */
608 sptlrpc_req_put_ctx(req, 0);
609 req->rq_cli_ctx = oldctx;
613 LASSERT(req->rq_cli_ctx == newctx);
616 sptlrpc_cli_ctx_put(oldctx, 1);
619 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
622 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
624 if (cli_ctx_is_refreshed(ctx))
630 void ctx_refresh_interrupt(struct ptlrpc_request *req)
633 spin_lock(&req->rq_lock);
635 spin_unlock(&req->rq_lock);
639 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
641 spin_lock(&ctx->cc_lock);
642 if (!list_empty(&req->rq_ctx_chain))
643 list_del_init(&req->rq_ctx_chain);
644 spin_unlock(&ctx->cc_lock);
648 * To refresh the context of \req, if it's not up-to-date.
650 * - == 0: do not wait
651 * - == MAX_SCHEDULE_TIMEOUT: wait indefinitely
652 * - > 0: not supported
654 * The status of the context could be subject to be changed by other threads
655 * at any time. We allow this race, but once we return with 0, the caller will
656 * suppose it's uptodated and keep using it until the owning rpc is done.
658 * \retval 0 only if the context is uptodated.
659 * \retval -ev error number.
661 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
663 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
664 struct ptlrpc_sec *sec;
671 if (req->rq_ctx_init || req->rq_ctx_fini)
674 if (timeout != 0 && timeout != MAX_SCHEDULE_TIMEOUT) {
675 CERROR("req %p: invalid timeout %lu\n", req, timeout);
680 * during the process a request's context might change type even
681 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
685 rc = import_sec_validate_get(req->rq_import, &sec);
689 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
690 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
691 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
692 req_off_ctx_list(req, ctx);
693 sptlrpc_req_replace_dead_ctx(req);
694 ctx = req->rq_cli_ctx;
696 sptlrpc_sec_put(sec);
698 if (cli_ctx_is_eternal(ctx))
701 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
702 if (ctx->cc_ops->refresh)
703 ctx->cc_ops->refresh(ctx);
705 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
707 LASSERT(ctx->cc_ops->validate);
708 if (ctx->cc_ops->validate(ctx) == 0) {
709 req_off_ctx_list(req, ctx);
713 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
714 spin_lock(&req->rq_lock);
716 spin_unlock(&req->rq_lock);
717 req_off_ctx_list(req, ctx);
722 * There's a subtle issue for resending RPCs, suppose following
724 * 1. the request was sent to server.
725 * 2. recovery was kicked start, after finished the request was
727 * 3. resend the request.
728 * 4. old reply from server received, we accept and verify the reply.
729 * this has to be success, otherwise the error will be aware
731 * 5. new reply from server received, dropped by LNet.
733 * Note the xid of old & new request is the same. We can't simply
734 * change xid for the resent request because the server replies on
735 * it for reply reconstruction.
737 * Commonly the original context should be uptodate because we
738 * have an expiry nice time; server will keep its context because
739 * we at least hold a ref of old context which prevent context
740 * from destroying RPC being sent. So server still can accept the
741 * request and finish the RPC. But if that's not the case:
742 * 1. If server side context has been trimmed, a NO_CONTEXT will
743 * be returned, gss_cli_ctx_verify/unseal will switch to new
745 * 2. Current context never be refreshed, then we are fine: we
746 * never really send request with old context before.
748 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
749 unlikely(req->rq_reqmsg) &&
750 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
751 req_off_ctx_list(req, ctx);
755 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
756 req_off_ctx_list(req, ctx);
758 * don't switch ctx if import was deactivated
760 if (req->rq_import->imp_deactive) {
761 spin_lock(&req->rq_lock);
763 spin_unlock(&req->rq_lock);
767 rc = sptlrpc_req_replace_dead_ctx(req);
769 LASSERT(ctx == req->rq_cli_ctx);
770 CERROR("req %p: failed to replace dead ctx %p: %d\n",
772 spin_lock(&req->rq_lock);
774 spin_unlock(&req->rq_lock);
778 ctx = req->rq_cli_ctx;
783 * Now we're sure this context is during upcall, add myself into
786 spin_lock(&ctx->cc_lock);
787 if (list_empty(&req->rq_ctx_chain))
788 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
789 spin_unlock(&ctx->cc_lock);
794 /* Clear any flags that may be present from previous sends */
795 LASSERT(req->rq_receiving_reply == 0);
796 spin_lock(&req->rq_lock);
798 req->rq_timedout = 0;
801 spin_unlock(&req->rq_lock);
803 /* by now we know that timeout value is MAX_SCHEDULE_TIMEOUT,
804 * so wait indefinitely with non-fatal signals blocked
806 if (l_wait_event_abortable(req->rq_reply_waitq,
807 ctx_check_refresh(ctx)) == -ERESTARTSYS) {
809 ctx_refresh_interrupt(req);
813 * following cases could lead us here:
814 * - successfully refreshed;
816 * - timedout, and we don't want recover from the failure;
817 * - timedout, and waked up upon recovery finished;
818 * - someone else mark this ctx dead by force;
819 * - someone invalidate the req and call ptlrpc_client_wake_req(),
820 * e.g. ptlrpc_abort_inflight();
822 if (!cli_ctx_is_refreshed(ctx)) {
823 /* timed out or interruptted */
824 req_off_ctx_list(req, ctx);
833 /* Bring ptlrpc_sec context up-to-date */
834 int sptlrpc_export_update_ctx(struct obd_export *exp)
836 struct obd_import *imp = exp ? exp->exp_imp_reverse : NULL;
837 struct ptlrpc_sec *sec = NULL;
838 struct ptlrpc_cli_ctx *ctx = NULL;
842 sec = sptlrpc_import_sec_ref(imp);
844 ctx = get_my_ctx(sec);
845 sptlrpc_sec_put(sec);
849 if (ctx->cc_ops->refresh)
850 rc = ctx->cc_ops->refresh(ctx);
851 sptlrpc_cli_ctx_put(ctx, 1);
857 * Initialize flavor settings for \a req, according to \a opcode.
859 * \note this could be called in two situations:
860 * - new request from ptlrpc_pre_req(), with proper @opcode
861 * - old request which changed ctx in the middle, with @opcode == 0
863 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
865 struct ptlrpc_sec *sec;
867 LASSERT(req->rq_import);
868 LASSERT(req->rq_cli_ctx);
869 LASSERT(req->rq_cli_ctx->cc_sec);
870 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
872 /* special security flags according to opcode */
876 case MGS_CONFIG_READ:
878 req->rq_bulk_read = 1;
882 req->rq_bulk_write = 1;
885 req->rq_ctx_init = 1;
888 req->rq_ctx_fini = 1;
891 /* init/fini rpc won't be resend, so can't be here */
892 LASSERT(req->rq_ctx_init == 0);
893 LASSERT(req->rq_ctx_fini == 0);
895 /* cleanup flags, which should be recalculated */
896 req->rq_pack_udesc = 0;
897 req->rq_pack_bulk = 0;
901 sec = req->rq_cli_ctx->cc_sec;
903 spin_lock(&sec->ps_lock);
904 req->rq_flvr = sec->ps_flvr;
905 spin_unlock(&sec->ps_lock);
908 * force SVC_NULL for context initiation rpc, SVC_INTG for context
911 if (unlikely(req->rq_ctx_init))
912 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
913 else if (unlikely(req->rq_ctx_fini))
914 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
916 /* user descriptor flag, null security can't do it anyway */
917 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
918 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
919 req->rq_pack_udesc = 1;
921 /* bulk security flag */
922 if ((req->rq_bulk_read || req->rq_bulk_write) &&
923 sptlrpc_flavor_has_bulk(&req->rq_flvr))
924 req->rq_pack_bulk = 1;
927 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
929 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
932 LASSERT(req->rq_clrbuf);
933 if (req->rq_pool || !req->rq_reqbuf)
936 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
937 req->rq_reqbuf = NULL;
938 req->rq_reqbuf_len = 0;
942 * Given an import \a imp, check whether current user has a valid context
943 * or not. We may create a new context and try to refresh it, and try
944 * repeatedly try in case of non-fatal errors. Return 0 means success.
946 int sptlrpc_import_check_ctx(struct obd_import *imp)
948 struct ptlrpc_sec *sec;
949 struct ptlrpc_cli_ctx *ctx;
950 struct ptlrpc_request *req = NULL;
957 sec = sptlrpc_import_sec_ref(imp);
958 ctx = get_my_ctx(sec);
959 sptlrpc_sec_put(sec);
964 if (cli_ctx_is_eternal(ctx) ||
965 ctx->cc_ops->validate(ctx) == 0) {
966 sptlrpc_cli_ctx_put(ctx, 1);
970 if (cli_ctx_is_error(ctx)) {
971 sptlrpc_cli_ctx_put(ctx, 1);
975 req = ptlrpc_request_cache_alloc(GFP_NOFS);
979 ptlrpc_cli_req_init(req);
980 atomic_set(&req->rq_refcount, 10000);
982 req->rq_import = imp;
983 req->rq_flvr = sec->ps_flvr;
984 req->rq_cli_ctx = ctx;
986 rc = sptlrpc_req_refresh_ctx(req, MAX_SCHEDULE_TIMEOUT);
987 LASSERT(list_empty(&req->rq_ctx_chain));
988 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
989 ptlrpc_request_cache_free(req);
995 * Used by ptlrpc client, to perform the pre-defined security transformation
996 * upon the request message of \a req. After this function called,
997 * req->rq_reqmsg is still accessible as clear text.
999 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1001 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1007 LASSERT(ctx->cc_sec);
1008 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1011 * we wrap bulk request here because now we can be sure
1012 * the context is uptodate.
1015 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1020 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1021 case SPTLRPC_SVC_NULL:
1022 case SPTLRPC_SVC_AUTH:
1023 case SPTLRPC_SVC_INTG:
1024 LASSERT(ctx->cc_ops->sign);
1025 rc = ctx->cc_ops->sign(ctx, req);
1027 case SPTLRPC_SVC_PRIV:
1028 LASSERT(ctx->cc_ops->seal);
1029 rc = ctx->cc_ops->seal(ctx, req);
1036 LASSERT(req->rq_reqdata_len);
1037 LASSERT(req->rq_reqdata_len % 8 == 0);
1038 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1044 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1046 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1052 LASSERT(ctx->cc_sec);
1053 LASSERT(req->rq_repbuf);
1054 LASSERT(req->rq_repdata);
1055 LASSERT(req->rq_repmsg == NULL);
1057 req->rq_rep_swab_mask = 0;
1059 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1062 req_capsule_set_rep_swabbed(&req->rq_pill,
1063 MSG_PTLRPC_HEADER_OFF);
1067 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1071 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1072 CERROR("replied data length %d too small\n",
1073 req->rq_repdata_len);
1077 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1078 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1079 CERROR("reply policy %u doesn't match request policy %u\n",
1080 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1081 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1085 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1086 case SPTLRPC_SVC_NULL:
1087 case SPTLRPC_SVC_AUTH:
1088 case SPTLRPC_SVC_INTG:
1089 LASSERT(ctx->cc_ops->verify);
1090 rc = ctx->cc_ops->verify(ctx, req);
1092 case SPTLRPC_SVC_PRIV:
1093 LASSERT(ctx->cc_ops->unseal);
1094 rc = ctx->cc_ops->unseal(ctx, req);
1099 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1101 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1103 req->rq_rep_swab_mask = 0;
1108 * Used by ptlrpc client, to perform security transformation upon the reply
1109 * message of \a req. After return successfully, req->rq_repmsg points to
1110 * the reply message in clear text.
1112 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1115 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1117 LASSERT(req->rq_repbuf);
1118 LASSERT(req->rq_repdata == NULL);
1119 LASSERT(req->rq_repmsg == NULL);
1120 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1122 if (req->rq_reply_off == 0 &&
1123 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1124 CERROR("real reply with offset 0\n");
1128 if (req->rq_reply_off % 8 != 0) {
1129 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1133 req->rq_repdata = (struct lustre_msg *)
1134 (req->rq_repbuf + req->rq_reply_off);
1135 req->rq_repdata_len = req->rq_nob_received;
1137 return do_cli_unwrap_reply(req);
1141 * Used by ptlrpc client, to perform security transformation upon the early
1142 * reply message of \a req. We expect the rq_reply_off is 0, and
1143 * rq_nob_received is the early reply size.
1145 * Because the receive buffer might be still posted, the reply data might be
1146 * changed at any time, no matter we're holding rq_lock or not. For this reason
1147 * we allocate a separate ptlrpc_request and reply buffer for early reply
1150 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1151 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1152 * \a *req_ret to release it.
1153 * \retval -ev error number, and \a req_ret will not be set.
1155 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1156 struct ptlrpc_request **req_ret)
1158 struct ptlrpc_request *early_req;
1160 int early_bufsz, early_size;
1165 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1166 if (early_req == NULL)
1169 ptlrpc_cli_req_init(early_req);
1171 early_size = req->rq_nob_received;
1172 early_bufsz = size_roundup_power2(early_size);
1173 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1174 if (early_buf == NULL)
1175 GOTO(err_req, rc = -ENOMEM);
1177 /* sanity checkings and copy data out, do it inside spinlock */
1178 spin_lock(&req->rq_lock);
1180 if (req->rq_replied) {
1181 spin_unlock(&req->rq_lock);
1182 GOTO(err_buf, rc = -EALREADY);
1185 LASSERT(req->rq_repbuf);
1186 LASSERT(req->rq_repdata == NULL);
1187 LASSERT(req->rq_repmsg == NULL);
1189 if (req->rq_reply_off != 0) {
1190 CERROR("early reply with offset %u\n", req->rq_reply_off);
1191 spin_unlock(&req->rq_lock);
1192 GOTO(err_buf, rc = -EPROTO);
1195 if (req->rq_nob_received != early_size) {
1196 /* even another early arrived the size should be the same */
1197 CERROR("data size has changed from %u to %u\n",
1198 early_size, req->rq_nob_received);
1199 spin_unlock(&req->rq_lock);
1200 GOTO(err_buf, rc = -EINVAL);
1203 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1204 CERROR("early reply length %d too small\n",
1205 req->rq_nob_received);
1206 spin_unlock(&req->rq_lock);
1207 GOTO(err_buf, rc = -EALREADY);
1210 memcpy(early_buf, req->rq_repbuf, early_size);
1211 spin_unlock(&req->rq_lock);
1213 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1214 early_req->rq_flvr = req->rq_flvr;
1215 early_req->rq_repbuf = early_buf;
1216 early_req->rq_repbuf_len = early_bufsz;
1217 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1218 early_req->rq_repdata_len = early_size;
1219 early_req->rq_early = 1;
1220 early_req->rq_reqmsg = req->rq_reqmsg;
1222 rc = do_cli_unwrap_reply(early_req);
1224 DEBUG_REQ(D_ADAPTTO, early_req,
1225 "unwrap early reply: rc = %d", rc);
1229 LASSERT(early_req->rq_repmsg);
1230 *req_ret = early_req;
1234 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1236 OBD_FREE_LARGE(early_buf, early_bufsz);
1238 ptlrpc_request_cache_free(early_req);
1243 * Used by ptlrpc client, to release a processed early reply \a early_req.
1245 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1247 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1249 LASSERT(early_req->rq_repbuf);
1250 LASSERT(early_req->rq_repdata);
1251 LASSERT(early_req->rq_repmsg);
1253 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1254 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1255 ptlrpc_request_cache_free(early_req);
1258 /**************************************************
1260 **************************************************/
1263 * "fixed" sec (e.g. null) use sec_id < 0
1265 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1267 int sptlrpc_get_next_secid(void)
1269 return atomic_inc_return(&sptlrpc_sec_id);
1271 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1274 * client side high-level security APIs
1277 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1278 int grace, int force)
1280 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1282 LASSERT(policy->sp_cops);
1283 LASSERT(policy->sp_cops->flush_ctx_cache);
1285 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1288 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1290 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1292 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1293 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1294 LASSERT(policy->sp_cops->destroy_sec);
1296 CDEBUG(D_SEC, "%s@%p: being destroyed\n", sec->ps_policy->sp_name, sec);
1298 policy->sp_cops->destroy_sec(sec);
1299 sptlrpc_policy_put(policy);
1302 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1304 sec_cop_destroy_sec(sec);
1306 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1308 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1310 LASSERT(atomic_read(&(sec)->ps_refcount) > 0);
1312 if (sec->ps_policy->sp_cops->kill_sec) {
1313 sec->ps_policy->sp_cops->kill_sec(sec);
1315 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1319 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1322 atomic_inc(&sec->ps_refcount);
1326 EXPORT_SYMBOL(sptlrpc_sec_get);
1328 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1331 LASSERT(atomic_read(&(sec)->ps_refcount) > 0);
1333 if (atomic_dec_and_test(&sec->ps_refcount)) {
1334 sptlrpc_gc_del_sec(sec);
1335 sec_cop_destroy_sec(sec);
1339 EXPORT_SYMBOL(sptlrpc_sec_put);
1342 * policy module is responsible for taking refrence of import
1345 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1346 struct ptlrpc_svc_ctx *svc_ctx,
1347 struct sptlrpc_flavor *sf,
1348 enum lustre_sec_part sp)
1350 struct ptlrpc_sec_policy *policy;
1351 struct ptlrpc_sec *sec;
1357 LASSERT(imp->imp_dlm_fake == 1);
1359 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1360 imp->imp_obd->obd_type->typ_name,
1361 imp->imp_obd->obd_name,
1362 sptlrpc_flavor2name(sf, str, sizeof(str)));
1364 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1365 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1367 LASSERT(imp->imp_dlm_fake == 0);
1369 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1370 imp->imp_obd->obd_type->typ_name,
1371 imp->imp_obd->obd_name,
1372 sptlrpc_flavor2name(sf, str, sizeof(str)));
1374 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1376 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1381 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1383 atomic_inc(&sec->ps_refcount);
1387 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1388 sptlrpc_gc_add_sec(sec);
1390 sptlrpc_policy_put(policy);
1396 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1398 struct ptlrpc_sec *sec;
1400 read_lock(&imp->imp_sec_lock);
1401 sec = sptlrpc_sec_get(imp->imp_sec);
1402 read_unlock(&imp->imp_sec_lock);
1406 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1408 static void sptlrpc_import_sec_install(struct obd_import *imp,
1409 struct ptlrpc_sec *sec)
1411 struct ptlrpc_sec *old_sec;
1413 LASSERT(atomic_read(&(sec)->ps_refcount) > 0);
1415 write_lock(&imp->imp_sec_lock);
1416 old_sec = imp->imp_sec;
1418 write_unlock(&imp->imp_sec_lock);
1421 sptlrpc_sec_kill(old_sec);
1423 /* balance the ref taken by this import */
1424 sptlrpc_sec_put(old_sec);
1429 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1431 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1435 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1441 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1442 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1444 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1445 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1447 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1448 struct ptlrpc_svc_ctx *svc_ctx,
1449 struct sptlrpc_flavor *flvr)
1451 struct ptlrpc_connection *conn;
1452 struct sptlrpc_flavor sf;
1453 struct ptlrpc_sec *sec, *newsec;
1454 enum lustre_sec_part sp;
1465 conn = imp->imp_connection;
1467 if (svc_ctx == NULL) {
1468 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1470 * normal import, determine flavor from rule set, except
1471 * for mgc the flavor is predetermined.
1473 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1474 sf = cliobd->cl_flvr_mgc;
1476 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1478 &cliobd->cl_target_uuid,
1479 &conn->c_self, &sf);
1481 sp = imp->imp_obd->u.cli.cl_sp_me;
1483 /* reverse import, determine flavor from incoming reqeust */
1486 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1487 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1488 PTLRPC_SEC_FL_ROOTONLY;
1490 sp = sptlrpc_target_sec_part(imp->imp_obd);
1493 sec = sptlrpc_import_sec_ref(imp);
1497 if (flavor_equal(&sf, &sec->ps_flvr))
1500 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1501 imp->imp_obd->obd_name,
1502 obd_uuid2str(&conn->c_remote_uuid),
1503 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1504 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1505 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1506 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1507 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1508 imp->imp_obd->obd_name,
1509 obd_uuid2str(&conn->c_remote_uuid),
1510 LNET_NID_NET(&conn->c_self),
1511 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1514 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1516 sptlrpc_import_sec_install(imp, newsec);
1518 CERROR("import %s->%s: failed to create new sec\n",
1519 imp->imp_obd->obd_name,
1520 obd_uuid2str(&conn->c_remote_uuid));
1525 sptlrpc_sec_put(sec);
1529 void sptlrpc_import_sec_put(struct obd_import *imp)
1532 sptlrpc_sec_kill(imp->imp_sec);
1534 sptlrpc_sec_put(imp->imp_sec);
1535 imp->imp_sec = NULL;
1539 static void import_flush_ctx_common(struct obd_import *imp,
1540 uid_t uid, int grace, int force)
1542 struct ptlrpc_sec *sec;
1547 sec = sptlrpc_import_sec_ref(imp);
1551 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1552 sptlrpc_sec_put(sec);
1555 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1558 * it's important to use grace mode, see explain in
1559 * sptlrpc_req_refresh_ctx()
1561 import_flush_ctx_common(imp, 0, 1, 1);
1564 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1566 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1569 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1571 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1573 import_flush_ctx_common(imp, -1, 1, 1);
1575 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1578 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1579 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1581 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1583 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1584 struct ptlrpc_sec_policy *policy;
1588 LASSERT(ctx->cc_sec);
1589 LASSERT(ctx->cc_sec->ps_policy);
1590 LASSERT(req->rq_reqmsg == NULL);
1591 LASSERT(atomic_read(&(ctx)->cc_refcount) > 0);
1593 policy = ctx->cc_sec->ps_policy;
1594 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1596 LASSERT(req->rq_reqmsg);
1597 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1599 /* zeroing preallocated buffer */
1601 memset(req->rq_reqmsg, 0, msgsize);
1608 * Used by ptlrpc client to free request buffer of \a req. After this
1609 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1611 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1613 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1614 struct ptlrpc_sec_policy *policy;
1617 LASSERT(ctx->cc_sec);
1618 LASSERT(ctx->cc_sec->ps_policy);
1619 LASSERT(atomic_read(&(ctx)->cc_refcount) > 0);
1621 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1624 policy = ctx->cc_sec->ps_policy;
1625 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1626 req->rq_reqmsg = NULL;
1630 * NOTE caller must guarantee the buffer size is enough for the enlargement
1632 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1633 int segment, int newsize)
1636 int oldsize, oldmsg_size, movesize;
1638 LASSERT(segment < msg->lm_bufcount);
1639 LASSERT(msg->lm_buflens[segment] <= newsize);
1641 if (msg->lm_buflens[segment] == newsize)
1644 /* nothing to do if we are enlarging the last segment */
1645 if (segment == msg->lm_bufcount - 1) {
1646 msg->lm_buflens[segment] = newsize;
1650 oldsize = msg->lm_buflens[segment];
1652 src = lustre_msg_buf(msg, segment + 1, 0);
1653 msg->lm_buflens[segment] = newsize;
1654 dst = lustre_msg_buf(msg, segment + 1, 0);
1655 msg->lm_buflens[segment] = oldsize;
1657 /* move from segment + 1 to end segment */
1658 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1659 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1660 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1661 LASSERT(movesize >= 0);
1664 memmove(dst, src, movesize);
1666 /* note we don't clear the ares where old data live, not secret */
1668 /* finally set new segment size */
1669 msg->lm_buflens[segment] = newsize;
1671 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1674 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1675 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1676 * preserved after the enlargement. this must be called after original request
1677 * buffer being allocated.
1679 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1680 * so caller should refresh its local pointers if needed.
1682 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1683 const struct req_msg_field *field,
1686 struct req_capsule *pill = &req->rq_pill;
1687 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1688 struct ptlrpc_sec_cops *cops;
1689 struct lustre_msg *msg = req->rq_reqmsg;
1690 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1694 LASSERT(msg->lm_bufcount > segment);
1695 LASSERT(msg->lm_buflens[segment] <= newsize);
1697 if (msg->lm_buflens[segment] == newsize)
1700 cops = ctx->cc_sec->ps_policy->sp_cops;
1701 LASSERT(cops->enlarge_reqbuf);
1702 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1704 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1707 * Used by ptlrpc client to allocate reply buffer of \a req.
1709 * \note After this, req->rq_repmsg is still not accessible.
1711 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1713 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1714 struct ptlrpc_sec_policy *policy;
1719 LASSERT(ctx->cc_sec);
1720 LASSERT(ctx->cc_sec->ps_policy);
1725 policy = ctx->cc_sec->ps_policy;
1726 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1730 * Used by ptlrpc client to free reply buffer of \a req. After this
1731 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1733 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1735 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1736 struct ptlrpc_sec_policy *policy;
1741 LASSERT(ctx->cc_sec);
1742 LASSERT(ctx->cc_sec->ps_policy);
1743 LASSERT(atomic_read(&(ctx)->cc_refcount) > 0);
1745 if (req->rq_repbuf == NULL)
1747 LASSERT(req->rq_repbuf_len);
1749 policy = ctx->cc_sec->ps_policy;
1750 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1751 req->rq_repmsg = NULL;
1754 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1756 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1757 struct ptlrpc_cli_ctx *ctx)
1759 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1761 if (!policy->sp_cops->install_rctx)
1763 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1766 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1767 struct ptlrpc_svc_ctx *ctx)
1769 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1771 if (!policy->sp_sops->install_rctx)
1773 return policy->sp_sops->install_rctx(imp, ctx);
1776 /* Get SELinux policy info from userspace */
1777 static int sepol_helper(struct obd_import *imp)
1779 char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1781 [0] = "/usr/sbin/l_getsepol",
1783 [2] = NULL, /* obd type */
1785 [4] = NULL, /* obd name */
1787 [6] = mtime_str, /* policy mtime */
1789 [8] = mode_str, /* enforcing mode */
1794 [1] = "PATH=/sbin:/usr/sbin",
1800 if (imp == NULL || imp->imp_obd == NULL ||
1801 imp->imp_obd->obd_type == NULL) {
1804 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1805 argv[4] = imp->imp_obd->obd_name;
1806 spin_lock(&imp->imp_sec->ps_lock);
1807 if (ktime_to_ns(imp->imp_sec->ps_sepol_mtime) == 0 &&
1808 imp->imp_sec->ps_sepol[0] == '\0') {
1809 /* ps_sepol has not been initialized */
1815 mtime_ms = ktime_to_ms(imp->imp_sec->ps_sepol_mtime);
1816 snprintf(mtime_str, sizeof(mtime_str), "%lld",
1817 mtime_ms / MSEC_PER_SEC);
1818 mode_str[0] = imp->imp_sec->ps_sepol[0];
1820 spin_unlock(&imp->imp_sec->ps_lock);
1821 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1828 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1832 if (send_sepol == 0)
1835 if (send_sepol == -1)
1836 /* send_sepol == -1 means fetch sepol status every time */
1839 spin_lock(&imp_sec->ps_lock);
1840 checknext = imp_sec->ps_sepol_checknext;
1841 spin_unlock(&imp_sec->ps_lock);
1843 /* next check is too far in time, please update */
1844 if (ktime_after(checknext,
1845 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1848 if (ktime_before(ktime_get(), checknext))
1849 /* too early to fetch sepol status */
1853 /* define new sepol_checknext time */
1854 spin_lock(&imp_sec->ps_lock);
1855 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1856 ktime_set(send_sepol, 0));
1857 spin_unlock(&imp_sec->ps_lock);
1862 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1864 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1869 (req->rq_sepol)[0] = '\0';
1871 #ifndef HAVE_SELINUX
1872 if (unlikely(send_sepol != 0))
1874 "Client cannot report SELinux status, it was not built against libselinux.\n");
1878 if (send_sepol == 0)
1881 if (imp_sec == NULL)
1884 /* Retrieve SELinux status info */
1885 if (sptlrpc_sepol_needs_check(imp_sec))
1886 rc = sepol_helper(req->rq_import);
1887 if (likely(rc == 0)) {
1888 spin_lock(&imp_sec->ps_lock);
1889 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1890 sizeof(req->rq_sepol));
1891 spin_unlock(&imp_sec->ps_lock);
1892 } else if (rc == -ENODEV) {
1894 "Client cannot report SELinux status, SELinux is disabled.\n");
1900 EXPORT_SYMBOL(sptlrpc_get_sepol);
1903 * server side security
1906 static int flavor_allowed(struct sptlrpc_flavor *exp,
1907 struct ptlrpc_request *req)
1909 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1911 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1914 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1915 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1916 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1917 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1923 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1926 * Given an export \a exp, check whether the flavor of incoming \a req
1927 * is allowed by the export \a exp. Main logic is about taking care of
1928 * changing configurations. Return 0 means success.
1930 int sptlrpc_target_export_check(struct obd_export *exp,
1931 struct ptlrpc_request *req)
1933 struct sptlrpc_flavor flavor;
1939 * client side export has no imp_reverse, skip
1940 * FIXME maybe we should check flavor this as well???
1942 if (exp->exp_imp_reverse == NULL)
1945 /* don't care about ctx fini rpc */
1946 if (req->rq_ctx_fini)
1949 spin_lock(&exp->exp_lock);
1952 * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1953 * the first req with the new flavor, then treat it as current flavor,
1954 * adapt reverse sec according to it.
1955 * note the first rpc with new flavor might not be with root ctx, in
1956 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1958 if (unlikely(exp->exp_flvr_changed) &&
1959 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1961 * make the new flavor as "current", and old ones as
1964 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1965 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1966 flavor = exp->exp_flvr_old[1];
1967 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1968 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1969 exp->exp_flvr_old[0] = exp->exp_flvr;
1970 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1971 EXP_FLVR_UPDATE_EXPIRE;
1972 exp->exp_flvr = flavor;
1974 /* flavor change finished */
1975 exp->exp_flvr_changed = 0;
1976 LASSERT(exp->exp_flvr_adapt == 1);
1978 /* if it's gss, we only interested in root ctx init */
1979 if (req->rq_auth_gss &&
1980 !(req->rq_ctx_init &&
1981 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1982 req->rq_auth_usr_ost))) {
1983 spin_unlock(&exp->exp_lock);
1984 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1985 req->rq_auth_gss, req->rq_ctx_init,
1986 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1987 req->rq_auth_usr_ost);
1991 exp->exp_flvr_adapt = 0;
1992 spin_unlock(&exp->exp_lock);
1994 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1995 req->rq_svc_ctx, &flavor);
1999 * if it equals to the current flavor, we accept it, but need to
2000 * dealing with reverse sec/ctx
2002 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
2004 * most cases should return here, we only interested in
2007 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2008 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2009 !req->rq_auth_usr_ost)) {
2010 spin_unlock(&exp->exp_lock);
2015 * if flavor just changed, we should not proceed, just leave
2016 * it and current flavor will be discovered and replaced
2017 * shortly, and let _this_ rpc pass through
2019 if (exp->exp_flvr_changed) {
2020 LASSERT(exp->exp_flvr_adapt);
2021 spin_unlock(&exp->exp_lock);
2025 if (exp->exp_flvr_adapt) {
2026 exp->exp_flvr_adapt = 0;
2027 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2028 exp, exp->exp_flvr.sf_rpc,
2029 exp->exp_flvr_old[0].sf_rpc,
2030 exp->exp_flvr_old[1].sf_rpc);
2031 flavor = exp->exp_flvr;
2032 spin_unlock(&exp->exp_lock);
2034 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2039 "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2040 exp, exp->exp_flvr.sf_rpc,
2041 exp->exp_flvr_old[0].sf_rpc,
2042 exp->exp_flvr_old[1].sf_rpc);
2043 spin_unlock(&exp->exp_lock);
2045 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2050 if (exp->exp_flvr_expire[0]) {
2051 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2052 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2054 "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2055 exp, exp->exp_flvr.sf_rpc,
2056 exp->exp_flvr_old[0].sf_rpc,
2057 exp->exp_flvr_old[1].sf_rpc,
2058 (s64)(exp->exp_flvr_expire[0] -
2059 ktime_get_real_seconds()));
2060 spin_unlock(&exp->exp_lock);
2064 CDEBUG(D_SEC, "mark middle expired\n");
2065 exp->exp_flvr_expire[0] = 0;
2067 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2068 exp->exp_flvr.sf_rpc,
2069 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2070 req->rq_flvr.sf_rpc);
2074 * now it doesn't match the current flavor, the only chance we can
2075 * accept it is match the old flavors which is not expired.
2077 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2078 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2079 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2080 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2082 exp->exp_flvr.sf_rpc,
2083 exp->exp_flvr_old[0].sf_rpc,
2084 exp->exp_flvr_old[1].sf_rpc,
2085 (s64)(exp->exp_flvr_expire[1] -
2086 ktime_get_real_seconds()));
2087 spin_unlock(&exp->exp_lock);
2091 CDEBUG(D_SEC, "mark oldest expired\n");
2092 exp->exp_flvr_expire[1] = 0;
2094 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2095 exp, exp->exp_flvr.sf_rpc,
2096 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2097 req->rq_flvr.sf_rpc);
2099 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2100 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2101 exp->exp_flvr_old[1].sf_rpc);
2104 spin_unlock(&exp->exp_lock);
2106 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2107 exp, exp->exp_obd->obd_name,
2108 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2109 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2110 req->rq_flvr.sf_rpc,
2111 exp->exp_flvr.sf_rpc,
2112 exp->exp_flvr_old[0].sf_rpc,
2113 exp->exp_flvr_expire[0] ?
2114 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2115 exp->exp_flvr_old[1].sf_rpc,
2116 exp->exp_flvr_expire[1] ?
2117 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2120 EXPORT_SYMBOL(sptlrpc_target_export_check);
2122 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2123 struct sptlrpc_rule_set *rset)
2125 struct obd_export *exp;
2126 struct sptlrpc_flavor new_flvr;
2130 spin_lock(&obd->obd_dev_lock);
2132 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2133 if (exp->exp_connection == NULL)
2137 * note if this export had just been updated flavor
2138 * (exp_flvr_changed == 1), this will override the
2141 spin_lock(&exp->exp_lock);
2142 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2143 &exp->exp_connection->c_peer.nid,
2145 if (exp->exp_flvr_changed ||
2146 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2147 exp->exp_flvr_old[1] = new_flvr;
2148 exp->exp_flvr_expire[1] = 0;
2149 exp->exp_flvr_changed = 1;
2150 exp->exp_flvr_adapt = 1;
2152 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2153 exp, sptlrpc_part2name(exp->exp_sp_peer),
2154 exp->exp_flvr.sf_rpc,
2155 exp->exp_flvr_old[1].sf_rpc);
2157 spin_unlock(&exp->exp_lock);
2160 spin_unlock(&obd->obd_dev_lock);
2162 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2164 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2166 /* peer's claim is unreliable unless gss is being used */
2167 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2170 switch (req->rq_sp_from) {
2172 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2173 /* The below message is checked in sanity-sec test_33 */
2174 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2175 svc_rc = SECSVC_DROP;
2179 if (!req->rq_auth_usr_mdt) {
2180 /* The below message is checked in sanity-sec test_33 */
2181 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2182 svc_rc = SECSVC_DROP;
2186 if (!req->rq_auth_usr_ost) {
2187 /* The below message is checked in sanity-sec test_33 */
2188 DEBUG_REQ(D_ERROR, req, "faked source OST");
2189 svc_rc = SECSVC_DROP;
2194 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2195 !req->rq_auth_usr_ost) {
2196 /* The below message is checked in sanity-sec test_33 */
2197 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2198 svc_rc = SECSVC_DROP;
2203 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2204 svc_rc = SECSVC_DROP;
2211 * Used by ptlrpc server, to perform transformation upon request message of
2212 * incoming \a req. This must be the first thing to do with an incoming
2213 * request in ptlrpc layer.
2215 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2216 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2217 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2218 * reply message has been prepared.
2219 * \retval SECSVC_DROP failed, this request should be dropped.
2221 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2223 struct ptlrpc_sec_policy *policy;
2224 struct lustre_msg *msg = req->rq_reqbuf;
2230 LASSERT(req->rq_reqmsg == NULL);
2231 LASSERT(req->rq_repmsg == NULL);
2232 LASSERT(req->rq_svc_ctx == NULL);
2234 req->rq_req_swab_mask = 0;
2236 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2239 req_capsule_set_req_swabbed(&req->rq_pill,
2240 MSG_PTLRPC_HEADER_OFF);
2244 CERROR("error unpacking request from %s x%llu\n",
2245 libcfs_idstr(&req->rq_peer), req->rq_xid);
2246 RETURN(SECSVC_DROP);
2249 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2250 req->rq_sp_from = LUSTRE_SP_ANY;
2251 req->rq_auth_uid = -1; /* set to INVALID_UID */
2252 req->rq_auth_mapped_uid = -1;
2254 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2256 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2257 RETURN(SECSVC_DROP);
2260 LASSERT(policy->sp_sops->accept);
2261 rc = policy->sp_sops->accept(req);
2262 sptlrpc_policy_put(policy);
2263 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2264 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2267 * if it's not null flavor (which means embedded packing msg),
2268 * reset the swab mask for the comming inner msg unpacking.
2270 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2271 req->rq_req_swab_mask = 0;
2273 /* sanity check for the request source */
2274 rc = sptlrpc_svc_check_from(req, rc);
2279 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2280 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2281 * a buffer of \a msglen size.
2283 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2285 struct ptlrpc_sec_policy *policy;
2286 struct ptlrpc_reply_state *rs;
2291 LASSERT(req->rq_svc_ctx);
2292 LASSERT(req->rq_svc_ctx->sc_policy);
2294 policy = req->rq_svc_ctx->sc_policy;
2295 LASSERT(policy->sp_sops->alloc_rs);
2297 rc = policy->sp_sops->alloc_rs(req, msglen);
2298 if (unlikely(rc == -ENOMEM)) {
2299 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2301 if (svcpt->scp_service->srv_max_reply_size <
2302 msglen + sizeof(struct ptlrpc_reply_state)) {
2303 /* Just return failure if the size is too big */
2304 CERROR("size of message is too big (%zd), %d allowed\n",
2305 msglen + sizeof(struct ptlrpc_reply_state),
2306 svcpt->scp_service->srv_max_reply_size);
2310 /* failed alloc, try emergency pool */
2311 rs = lustre_get_emerg_rs(svcpt);
2315 req->rq_reply_state = rs;
2316 rc = policy->sp_sops->alloc_rs(req, msglen);
2318 lustre_put_emerg_rs(rs);
2319 req->rq_reply_state = NULL;
2324 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2330 * Used by ptlrpc server, to perform transformation upon reply message.
2332 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2333 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2335 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2337 struct ptlrpc_sec_policy *policy;
2342 LASSERT(req->rq_svc_ctx);
2343 LASSERT(req->rq_svc_ctx->sc_policy);
2345 policy = req->rq_svc_ctx->sc_policy;
2346 LASSERT(policy->sp_sops->authorize);
2348 rc = policy->sp_sops->authorize(req);
2349 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2355 * Used by ptlrpc server, to free reply_state.
2357 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2359 struct ptlrpc_sec_policy *policy;
2360 unsigned int prealloc;
2364 LASSERT(rs->rs_svc_ctx);
2365 LASSERT(rs->rs_svc_ctx->sc_policy);
2367 policy = rs->rs_svc_ctx->sc_policy;
2368 LASSERT(policy->sp_sops->free_rs);
2370 prealloc = rs->rs_prealloc;
2371 policy->sp_sops->free_rs(rs);
2374 lustre_put_emerg_rs(rs);
2378 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2380 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2383 atomic_inc(&ctx->sc_refcount);
2386 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2388 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2393 LASSERT(atomic_read(&(ctx)->sc_refcount) > 0);
2394 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2395 if (ctx->sc_policy->sp_sops->free_ctx)
2396 ctx->sc_policy->sp_sops->free_ctx(ctx);
2398 req->rq_svc_ctx = NULL;
2401 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2403 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2408 LASSERT(atomic_read(&(ctx)->sc_refcount) > 0);
2409 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2410 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2412 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2419 * Perform transformation upon bulk data pointed by \a desc. This is called
2420 * before transforming the request message.
2422 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2423 struct ptlrpc_bulk_desc *desc)
2425 struct ptlrpc_cli_ctx *ctx;
2427 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2429 if (!req->rq_pack_bulk)
2432 ctx = req->rq_cli_ctx;
2433 if (ctx->cc_ops->wrap_bulk)
2434 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2437 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2440 * This is called after unwrap the reply message.
2441 * return nob of actual plain text size received, or error code.
2443 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2444 struct ptlrpc_bulk_desc *desc,
2447 struct ptlrpc_cli_ctx *ctx;
2450 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2452 if (!req->rq_pack_bulk)
2453 return desc->bd_nob_transferred;
2455 ctx = req->rq_cli_ctx;
2456 if (ctx->cc_ops->unwrap_bulk) {
2457 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2461 return desc->bd_nob_transferred;
2463 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2466 * This is called after unwrap the reply message.
2467 * return 0 for success or error code.
2469 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2470 struct ptlrpc_bulk_desc *desc)
2472 struct ptlrpc_cli_ctx *ctx;
2475 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2477 if (!req->rq_pack_bulk)
2480 ctx = req->rq_cli_ctx;
2481 if (ctx->cc_ops->unwrap_bulk) {
2482 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2488 * if everything is going right, nob should equals to nob_transferred.
2489 * in case of privacy mode, nob_transferred needs to be adjusted.
2491 if (desc->bd_nob != desc->bd_nob_transferred) {
2492 CERROR("nob %d doesn't match transferred nob %d\n",
2493 desc->bd_nob, desc->bd_nob_transferred);
2499 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2501 #ifdef HAVE_SERVER_SUPPORT
2503 * Performe transformation upon outgoing bulk read.
2505 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2506 struct ptlrpc_bulk_desc *desc)
2508 struct ptlrpc_svc_ctx *ctx;
2510 LASSERT(req->rq_bulk_read);
2512 if (!req->rq_pack_bulk)
2515 ctx = req->rq_svc_ctx;
2516 if (ctx->sc_policy->sp_sops->wrap_bulk)
2517 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2521 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2524 * Performe transformation upon incoming bulk write.
2526 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2527 struct ptlrpc_bulk_desc *desc)
2529 struct ptlrpc_svc_ctx *ctx;
2532 LASSERT(req->rq_bulk_write);
2535 * if it's in privacy mode, transferred should >= expected; otherwise
2536 * transferred should == expected.
2538 if (desc->bd_nob_transferred < desc->bd_nob ||
2539 (desc->bd_nob_transferred > desc->bd_nob &&
2540 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2541 SPTLRPC_BULK_SVC_PRIV)) {
2542 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2543 desc->bd_nob_transferred, desc->bd_nob);
2547 if (!req->rq_pack_bulk)
2550 ctx = req->rq_svc_ctx;
2551 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2552 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2554 CERROR("error unwrap bulk: %d\n", rc);
2557 /* return 0 to allow reply be sent */
2560 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2563 * Prepare buffers for incoming bulk write.
2565 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2566 struct ptlrpc_bulk_desc *desc)
2568 struct ptlrpc_svc_ctx *ctx;
2570 LASSERT(req->rq_bulk_write);
2572 if (!req->rq_pack_bulk)
2575 ctx = req->rq_svc_ctx;
2576 if (ctx->sc_policy->sp_sops->prep_bulk)
2577 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2581 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2583 #endif /* HAVE_SERVER_SUPPORT */
2586 * user descriptor helpers
2589 int sptlrpc_current_user_desc_size(void)
2593 ngroups = current_cred()->group_info->ngroups;
2595 if (ngroups > LUSTRE_MAX_GROUPS)
2596 ngroups = LUSTRE_MAX_GROUPS;
2597 return sptlrpc_user_desc_size(ngroups);
2599 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2601 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2603 struct ptlrpc_user_desc *pud;
2606 pud = lustre_msg_buf(msg, offset, 0);
2608 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2609 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2610 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2611 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2612 pud->pud_cap = current_cap().cap[0];
2613 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2616 ngroups = current_cred()->group_info->ngroups;
2617 if (pud->pud_ngroups > ngroups)
2618 pud->pud_ngroups = ngroups;
2619 #ifdef HAVE_GROUP_INFO_GID
2620 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2621 pud->pud_ngroups * sizeof(__u32));
2622 #else /* !HAVE_GROUP_INFO_GID */
2623 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2624 pud->pud_ngroups * sizeof(__u32));
2625 #endif /* HAVE_GROUP_INFO_GID */
2626 task_unlock(current);
2630 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2632 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2634 struct ptlrpc_user_desc *pud;
2637 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2642 __swab32s(&pud->pud_uid);
2643 __swab32s(&pud->pud_gid);
2644 __swab32s(&pud->pud_fsuid);
2645 __swab32s(&pud->pud_fsgid);
2646 __swab32s(&pud->pud_cap);
2647 __swab32s(&pud->pud_ngroups);
2650 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2651 CERROR("%u groups is too large\n", pud->pud_ngroups);
2655 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2656 msg->lm_buflens[offset]) {
2657 CERROR("%u groups are claimed but bufsize only %u\n",
2658 pud->pud_ngroups, msg->lm_buflens[offset]);
2663 for (i = 0; i < pud->pud_ngroups; i++)
2664 __swab32s(&pud->pud_groups[i]);
2669 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2675 const char *sec2target_str(struct ptlrpc_sec *sec)
2677 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2679 if (sec_is_reverse(sec))
2681 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2683 EXPORT_SYMBOL(sec2target_str);
2686 * return true if the bulk data is protected
2688 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2690 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2691 case SPTLRPC_BULK_SVC_INTG:
2692 case SPTLRPC_BULK_SVC_PRIV:
2698 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2701 * crypto API helper/alloc blkciper
2705 * initialize/finalize
2708 int sptlrpc_init(void)
2712 rwlock_init(&policy_lock);
2714 rc = sptlrpc_gc_init();
2718 rc = sptlrpc_conf_init();
2722 rc = sptlrpc_enc_pool_init();
2726 rc = sptlrpc_null_init();
2730 rc = sptlrpc_plain_init();
2734 rc = sptlrpc_lproc_init();
2741 sptlrpc_plain_fini();
2743 sptlrpc_null_fini();
2745 sptlrpc_enc_pool_fini();
2747 sptlrpc_conf_fini();
2754 void sptlrpc_fini(void)
2756 sptlrpc_lproc_fini();
2757 sptlrpc_plain_fini();
2758 sptlrpc_null_fini();
2759 sptlrpc_enc_pool_fini();
2760 sptlrpc_conf_fini();