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, 2014, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Eric Mei <ericm@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_SEC
39 #include <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <linux/crypto.h>
44 #include <linux/key.h>
46 #include <libcfs/libcfs.h>
48 #include <obd_class.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
51 #include <lustre_import.h>
52 #include <lustre_dlm.h>
53 #include <lustre_sec.h>
55 #include "ptlrpc_internal.h"
57 /***********************************************
59 ***********************************************/
61 static rwlock_t policy_lock;
62 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
66 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
68 __u16 number = policy->sp_policy;
70 LASSERT(policy->sp_name);
71 LASSERT(policy->sp_cops);
72 LASSERT(policy->sp_sops);
74 if (number >= SPTLRPC_POLICY_MAX)
77 write_lock(&policy_lock);
78 if (unlikely(policies[number])) {
79 write_unlock(&policy_lock);
82 policies[number] = policy;
83 write_unlock(&policy_lock);
85 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
88 EXPORT_SYMBOL(sptlrpc_register_policy);
90 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
92 __u16 number = policy->sp_policy;
94 LASSERT(number < SPTLRPC_POLICY_MAX);
96 write_lock(&policy_lock);
97 if (unlikely(policies[number] == NULL)) {
98 write_unlock(&policy_lock);
99 CERROR("%s: already unregistered\n", policy->sp_name);
103 LASSERT(policies[number] == policy);
104 policies[number] = NULL;
105 write_unlock(&policy_lock);
107 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
110 EXPORT_SYMBOL(sptlrpc_unregister_policy);
113 struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
115 static DEFINE_MUTEX(load_mutex);
116 static atomic_t loaded = ATOMIC_INIT(0);
117 struct ptlrpc_sec_policy *policy;
118 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
121 if (number >= SPTLRPC_POLICY_MAX)
125 read_lock(&policy_lock);
126 policy = policies[number];
127 if (policy && !try_module_get(policy->sp_owner))
130 flag = atomic_read(&loaded);
131 read_unlock(&policy_lock);
133 if (policy != NULL || flag != 0 ||
134 number != SPTLRPC_POLICY_GSS)
137 /* try to load gss module, once */
138 mutex_lock(&load_mutex);
139 if (atomic_read(&loaded) == 0) {
140 if (request_module("ptlrpc_gss") == 0)
142 "module ptlrpc_gss loaded on demand\n");
144 CERROR("Unable to load module ptlrpc_gss\n");
146 atomic_set(&loaded, 1);
148 mutex_unlock(&load_mutex);
154 __u32 sptlrpc_name2flavor_base(const char *name)
156 if (!strcmp(name, "null"))
157 return SPTLRPC_FLVR_NULL;
158 if (!strcmp(name, "plain"))
159 return SPTLRPC_FLVR_PLAIN;
160 if (!strcmp(name, "gssnull"))
161 return SPTLRPC_FLVR_GSSNULL;
162 if (!strcmp(name, "krb5n"))
163 return SPTLRPC_FLVR_KRB5N;
164 if (!strcmp(name, "krb5a"))
165 return SPTLRPC_FLVR_KRB5A;
166 if (!strcmp(name, "krb5i"))
167 return SPTLRPC_FLVR_KRB5I;
168 if (!strcmp(name, "krb5p"))
169 return SPTLRPC_FLVR_KRB5P;
170 if (!strcmp(name, "skn"))
171 return SPTLRPC_FLVR_SKN;
172 if (!strcmp(name, "ska"))
173 return SPTLRPC_FLVR_SKA;
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_SKN))
203 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
205 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
207 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
210 CERROR("invalid wire flavor 0x%x\n", flvr);
213 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
215 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
216 char *buf, int bufsize)
218 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
219 snprintf(buf, bufsize, "hash:%s",
220 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
222 snprintf(buf, bufsize, "%s",
223 sptlrpc_flavor2name_base(sf->sf_rpc));
225 buf[bufsize - 1] = '\0';
228 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
230 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
232 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
235 * currently we don't support customized bulk specification for
236 * flavors other than plain
238 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
242 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
243 strncat(buf, bspec, bufsize);
246 buf[bufsize - 1] = '\0';
249 EXPORT_SYMBOL(sptlrpc_flavor2name);
251 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
255 if (flags & PTLRPC_SEC_FL_REVERSE)
256 strlcat(buf, "reverse,", bufsize);
257 if (flags & PTLRPC_SEC_FL_ROOTONLY)
258 strlcat(buf, "rootonly,", bufsize);
259 if (flags & PTLRPC_SEC_FL_UDESC)
260 strlcat(buf, "udesc,", bufsize);
261 if (flags & PTLRPC_SEC_FL_BULK)
262 strlcat(buf, "bulk,", bufsize);
264 strlcat(buf, "-,", bufsize);
268 EXPORT_SYMBOL(sptlrpc_secflags2str);
270 /**************************************************
271 * client context APIs *
272 **************************************************/
275 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
277 struct vfs_cred vcred;
278 int create = 1, remove_dead = 1;
281 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
283 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
284 PTLRPC_SEC_FL_ROOTONLY)) {
287 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
292 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
293 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
296 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
300 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
302 atomic_inc(&ctx->cc_refcount);
305 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
307 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
309 struct ptlrpc_sec *sec = ctx->cc_sec;
312 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
314 if (!atomic_dec_and_test(&ctx->cc_refcount))
317 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
319 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
322 * Expire the client context immediately.
324 * \pre Caller must hold at least 1 reference on the \a ctx.
326 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
328 LASSERT(ctx->cc_ops->die);
329 ctx->cc_ops->die(ctx, 0);
331 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
334 * To wake up the threads who are waiting for this client context. Called
335 * after some status change happened on \a ctx.
337 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
339 struct ptlrpc_request *req, *next;
341 spin_lock(&ctx->cc_lock);
342 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
344 list_del_init(&req->rq_ctx_chain);
345 ptlrpc_client_wake_req(req);
347 spin_unlock(&ctx->cc_lock);
349 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
351 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
353 LASSERT(ctx->cc_ops);
355 if (ctx->cc_ops->display == NULL)
358 return ctx->cc_ops->display(ctx, buf, bufsize);
361 static int import_sec_check_expire(struct obd_import *imp)
365 spin_lock(&imp->imp_lock);
366 if (imp->imp_sec_expire &&
367 imp->imp_sec_expire < ktime_get_real_seconds()) {
369 imp->imp_sec_expire = 0;
371 spin_unlock(&imp->imp_lock);
376 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
377 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
381 * Get and validate the client side ptlrpc security facilities from
382 * \a imp. There is a race condition on client reconnect when the import is
383 * being destroyed while there are outstanding client bound requests. In
384 * this case do not output any error messages if import secuity is not
387 * \param[in] imp obd import associated with client
388 * \param[out] sec client side ptlrpc security
390 * \retval 0 if security retrieved successfully
391 * \retval -ve errno if there was a problem
393 static int import_sec_validate_get(struct obd_import *imp,
394 struct ptlrpc_sec **sec)
398 if (unlikely(imp->imp_sec_expire)) {
399 rc = import_sec_check_expire(imp);
404 *sec = sptlrpc_import_sec_ref(imp);
405 /* Only output an error when the import is still active */
407 if (list_empty(&imp->imp_zombie_chain))
408 CERROR("import %p (%s) with no sec\n",
409 imp, ptlrpc_import_state_name(imp->imp_state));
413 if (unlikely((*sec)->ps_dying)) {
414 CERROR("attempt to use dying sec %p\n", sec);
415 sptlrpc_sec_put(*sec);
423 * Given a \a req, find or allocate an appropriate context for it.
424 * \pre req->rq_cli_ctx == NULL.
426 * \retval 0 succeed, and req->rq_cli_ctx is set.
427 * \retval -ev error number, and req->rq_cli_ctx == NULL.
429 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
431 struct obd_import *imp = req->rq_import;
432 struct ptlrpc_sec *sec;
436 LASSERT(!req->rq_cli_ctx);
439 rc = import_sec_validate_get(imp, &sec);
443 req->rq_cli_ctx = get_my_ctx(sec);
445 sptlrpc_sec_put(sec);
447 if (!req->rq_cli_ctx) {
448 CERROR("req %p: fail to get context\n", req);
449 RETURN(-ECONNREFUSED);
456 * Drop the context for \a req.
457 * \pre req->rq_cli_ctx != NULL.
458 * \post req->rq_cli_ctx == NULL.
460 * If \a sync == 0, this function should return quickly without sleep;
461 * otherwise it might trigger and wait for the whole process of sending
462 * an context-destroying rpc to server.
464 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
469 LASSERT(req->rq_cli_ctx);
471 /* request might be asked to release earlier while still
472 * in the context waiting list.
474 if (!list_empty(&req->rq_ctx_chain)) {
475 spin_lock(&req->rq_cli_ctx->cc_lock);
476 list_del_init(&req->rq_ctx_chain);
477 spin_unlock(&req->rq_cli_ctx->cc_lock);
480 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
481 req->rq_cli_ctx = NULL;
486 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
487 struct ptlrpc_cli_ctx *oldctx,
488 struct ptlrpc_cli_ctx *newctx)
490 struct sptlrpc_flavor old_flvr;
491 char *reqmsg = NULL; /* to workaround old gcc */
495 LASSERT(req->rq_reqmsg);
496 LASSERT(req->rq_reqlen);
497 LASSERT(req->rq_replen);
499 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
500 "switch sec %p(%s) -> %p(%s)\n", req,
501 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
502 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
503 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
504 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
507 old_flvr = req->rq_flvr;
509 /* save request message */
510 reqmsg_size = req->rq_reqlen;
511 if (reqmsg_size != 0) {
512 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
515 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
518 /* release old req/rep buf */
519 req->rq_cli_ctx = oldctx;
520 sptlrpc_cli_free_reqbuf(req);
521 sptlrpc_cli_free_repbuf(req);
522 req->rq_cli_ctx = newctx;
524 /* recalculate the flavor */
525 sptlrpc_req_set_flavor(req, 0);
527 /* alloc new request buffer
528 * we don't need to alloc reply buffer here, leave it to the
529 * rest procedure of ptlrpc */
530 if (reqmsg_size != 0) {
531 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
533 LASSERT(req->rq_reqmsg);
534 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
536 CWARN("failed to alloc reqbuf: %d\n", rc);
537 req->rq_flvr = old_flvr;
540 OBD_FREE_LARGE(reqmsg, reqmsg_size);
546 * If current context of \a req is dead somehow, e.g. we just switched flavor
547 * thus marked original contexts dead, we'll find a new context for it. if
548 * no switch is needed, \a req will end up with the same context.
550 * \note a request must have a context, to keep other parts of code happy.
551 * In any case of failure during the switching, we must restore the old one.
553 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
555 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
556 struct ptlrpc_cli_ctx *newctx;
562 sptlrpc_cli_ctx_get(oldctx);
563 sptlrpc_req_put_ctx(req, 0);
565 rc = sptlrpc_req_get_ctx(req);
567 LASSERT(!req->rq_cli_ctx);
569 /* restore old ctx */
570 req->rq_cli_ctx = oldctx;
574 newctx = req->rq_cli_ctx;
577 if (unlikely(newctx == oldctx &&
578 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
580 * still get the old dead ctx, usually means system too busy
583 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
584 newctx, newctx->cc_flags);
586 set_current_state(TASK_INTERRUPTIBLE);
587 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
588 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
591 * new ctx not up to date yet
594 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
595 newctx, newctx->cc_flags);
598 * it's possible newctx == oldctx if we're switching
599 * subflavor with the same sec.
601 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
603 /* restore old ctx */
604 sptlrpc_req_put_ctx(req, 0);
605 req->rq_cli_ctx = oldctx;
609 LASSERT(req->rq_cli_ctx == newctx);
612 sptlrpc_cli_ctx_put(oldctx, 1);
615 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
618 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
620 if (cli_ctx_is_refreshed(ctx))
626 int ctx_refresh_timeout(void *data)
628 struct ptlrpc_request *req = data;
631 /* conn_cnt is needed in expire_one_request */
632 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
634 rc = ptlrpc_expire_one_request(req, 1);
635 /* if we started recovery, we should mark this ctx dead; otherwise
636 * in case of lgssd died nobody would retire this ctx, following
637 * connecting will still find the same ctx thus cause deadlock.
638 * there's an assumption that expire time of the request should be
639 * later than the context refresh expire time.
642 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
647 void ctx_refresh_interrupt(void *data)
649 struct ptlrpc_request *req = data;
651 spin_lock(&req->rq_lock);
653 spin_unlock(&req->rq_lock);
657 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
659 spin_lock(&ctx->cc_lock);
660 if (!list_empty(&req->rq_ctx_chain))
661 list_del_init(&req->rq_ctx_chain);
662 spin_unlock(&ctx->cc_lock);
666 * To refresh the context of \req, if it's not up-to-date.
669 * - = 0: wait until success or fatal error occur
670 * - > 0: timeout value (in seconds)
672 * The status of the context could be subject to be changed by other threads
673 * at any time. We allow this race, but once we return with 0, the caller will
674 * suppose it's uptodated and keep using it until the owning rpc is done.
676 * \retval 0 only if the context is uptodated.
677 * \retval -ev error number.
679 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
681 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
682 struct ptlrpc_sec *sec;
683 struct l_wait_info lwi;
689 if (req->rq_ctx_init || req->rq_ctx_fini)
693 * during the process a request's context might change type even
694 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
698 rc = import_sec_validate_get(req->rq_import, &sec);
702 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
703 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
704 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
705 req_off_ctx_list(req, ctx);
706 sptlrpc_req_replace_dead_ctx(req);
707 ctx = req->rq_cli_ctx;
709 sptlrpc_sec_put(sec);
711 if (cli_ctx_is_eternal(ctx))
714 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
715 LASSERT(ctx->cc_ops->refresh);
716 ctx->cc_ops->refresh(ctx);
718 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
720 LASSERT(ctx->cc_ops->validate);
721 if (ctx->cc_ops->validate(ctx) == 0) {
722 req_off_ctx_list(req, ctx);
726 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
727 spin_lock(&req->rq_lock);
729 spin_unlock(&req->rq_lock);
730 req_off_ctx_list(req, ctx);
735 * There's a subtle issue for resending RPCs, suppose following
737 * 1. the request was sent to server.
738 * 2. recovery was kicked start, after finished the request was
740 * 3. resend the request.
741 * 4. old reply from server received, we accept and verify the reply.
742 * this has to be success, otherwise the error will be aware
744 * 5. new reply from server received, dropped by LNet.
746 * Note the xid of old & new request is the same. We can't simply
747 * change xid for the resent request because the server replies on
748 * it for reply reconstruction.
750 * Commonly the original context should be uptodate because we
751 * have an expiry nice time; server will keep its context because
752 * we at least hold a ref of old context which prevent context
753 * from destroying RPC being sent. So server still can accept the
754 * request and finish the RPC. But if that's not the case:
755 * 1. If server side context has been trimmed, a NO_CONTEXT will
756 * be returned, gss_cli_ctx_verify/unseal will switch to new
758 * 2. Current context never be refreshed, then we are fine: we
759 * never really send request with old context before.
761 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
762 unlikely(req->rq_reqmsg) &&
763 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
764 req_off_ctx_list(req, ctx);
768 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
769 req_off_ctx_list(req, ctx);
771 * don't switch ctx if import was deactivated
773 if (req->rq_import->imp_deactive) {
774 spin_lock(&req->rq_lock);
776 spin_unlock(&req->rq_lock);
780 rc = sptlrpc_req_replace_dead_ctx(req);
782 LASSERT(ctx == req->rq_cli_ctx);
783 CERROR("req %p: failed to replace dead ctx %p: %d\n",
785 spin_lock(&req->rq_lock);
787 spin_unlock(&req->rq_lock);
791 ctx = req->rq_cli_ctx;
796 * Now we're sure this context is during upcall, add myself into
799 spin_lock(&ctx->cc_lock);
800 if (list_empty(&req->rq_ctx_chain))
801 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
802 spin_unlock(&ctx->cc_lock);
805 RETURN(-EWOULDBLOCK);
807 /* Clear any flags that may be present from previous sends */
808 LASSERT(req->rq_receiving_reply == 0);
809 spin_lock(&req->rq_lock);
811 req->rq_timedout = 0;
814 spin_unlock(&req->rq_lock);
816 lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
818 ctx_refresh_interrupt, req);
819 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
822 * following cases could lead us here:
823 * - successfully refreshed;
825 * - timedout, and we don't want recover from the failure;
826 * - timedout, and waked up upon recovery finished;
827 * - someone else mark this ctx dead by force;
828 * - someone invalidate the req and call ptlrpc_client_wake_req(),
829 * e.g. ptlrpc_abort_inflight();
831 if (!cli_ctx_is_refreshed(ctx)) {
832 /* timed out or interruptted */
833 req_off_ctx_list(req, ctx);
843 * Initialize flavor settings for \a req, according to \a opcode.
845 * \note this could be called in two situations:
846 * - new request from ptlrpc_pre_req(), with proper @opcode
847 * - old request which changed ctx in the middle, with @opcode == 0
849 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
851 struct ptlrpc_sec *sec;
853 LASSERT(req->rq_import);
854 LASSERT(req->rq_cli_ctx);
855 LASSERT(req->rq_cli_ctx->cc_sec);
856 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
858 /* special security flags according to opcode */
862 case MGS_CONFIG_READ:
864 req->rq_bulk_read = 1;
868 req->rq_bulk_write = 1;
871 req->rq_ctx_init = 1;
874 req->rq_ctx_fini = 1;
877 /* init/fini rpc won't be resend, so can't be here */
878 LASSERT(req->rq_ctx_init == 0);
879 LASSERT(req->rq_ctx_fini == 0);
881 /* cleanup flags, which should be recalculated */
882 req->rq_pack_udesc = 0;
883 req->rq_pack_bulk = 0;
887 sec = req->rq_cli_ctx->cc_sec;
889 spin_lock(&sec->ps_lock);
890 req->rq_flvr = sec->ps_flvr;
891 spin_unlock(&sec->ps_lock);
893 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
895 if (unlikely(req->rq_ctx_init))
896 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
897 else if (unlikely(req->rq_ctx_fini))
898 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
900 /* user descriptor flag, null security can't do it anyway */
901 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
902 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
903 req->rq_pack_udesc = 1;
905 /* bulk security flag */
906 if ((req->rq_bulk_read || req->rq_bulk_write) &&
907 sptlrpc_flavor_has_bulk(&req->rq_flvr))
908 req->rq_pack_bulk = 1;
911 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
913 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
916 LASSERT(req->rq_clrbuf);
917 if (req->rq_pool || !req->rq_reqbuf)
920 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
921 req->rq_reqbuf = NULL;
922 req->rq_reqbuf_len = 0;
926 * Given an import \a imp, check whether current user has a valid context
927 * or not. We may create a new context and try to refresh it, and try
928 * repeatedly try in case of non-fatal errors. Return 0 means success.
930 int sptlrpc_import_check_ctx(struct obd_import *imp)
932 struct ptlrpc_sec *sec;
933 struct ptlrpc_cli_ctx *ctx;
934 struct ptlrpc_request *req = NULL;
940 sec = sptlrpc_import_sec_ref(imp);
941 ctx = get_my_ctx(sec);
942 sptlrpc_sec_put(sec);
947 if (cli_ctx_is_eternal(ctx) ||
948 ctx->cc_ops->validate(ctx) == 0) {
949 sptlrpc_cli_ctx_put(ctx, 1);
953 if (cli_ctx_is_error(ctx)) {
954 sptlrpc_cli_ctx_put(ctx, 1);
958 req = ptlrpc_request_cache_alloc(GFP_NOFS);
962 ptlrpc_cli_req_init(req);
963 atomic_set(&req->rq_refcount, 10000);
965 req->rq_import = imp;
966 req->rq_flvr = sec->ps_flvr;
967 req->rq_cli_ctx = ctx;
969 rc = sptlrpc_req_refresh_ctx(req, 0);
970 LASSERT(list_empty(&req->rq_ctx_chain));
971 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
972 ptlrpc_request_cache_free(req);
978 * Used by ptlrpc client, to perform the pre-defined security transformation
979 * upon the request message of \a req. After this function called,
980 * req->rq_reqmsg is still accessible as clear text.
982 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
984 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
989 LASSERT(ctx->cc_sec);
990 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
992 /* we wrap bulk request here because now we can be sure
993 * the context is uptodate.
996 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1001 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1002 case SPTLRPC_SVC_NULL:
1003 case SPTLRPC_SVC_AUTH:
1004 case SPTLRPC_SVC_INTG:
1005 LASSERT(ctx->cc_ops->sign);
1006 rc = ctx->cc_ops->sign(ctx, req);
1008 case SPTLRPC_SVC_PRIV:
1009 LASSERT(ctx->cc_ops->seal);
1010 rc = ctx->cc_ops->seal(ctx, req);
1017 LASSERT(req->rq_reqdata_len);
1018 LASSERT(req->rq_reqdata_len % 8 == 0);
1019 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1025 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1027 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1032 LASSERT(ctx->cc_sec);
1033 LASSERT(req->rq_repbuf);
1034 LASSERT(req->rq_repdata);
1035 LASSERT(req->rq_repmsg == NULL);
1037 req->rq_rep_swab_mask = 0;
1039 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1042 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1046 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1050 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1051 CERROR("replied data length %d too small\n",
1052 req->rq_repdata_len);
1056 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1057 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1058 CERROR("reply policy %u doesn't match request policy %u\n",
1059 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1060 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1064 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1065 case SPTLRPC_SVC_NULL:
1066 case SPTLRPC_SVC_AUTH:
1067 case SPTLRPC_SVC_INTG:
1068 LASSERT(ctx->cc_ops->verify);
1069 rc = ctx->cc_ops->verify(ctx, req);
1071 case SPTLRPC_SVC_PRIV:
1072 LASSERT(ctx->cc_ops->unseal);
1073 rc = ctx->cc_ops->unseal(ctx, req);
1078 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1080 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1082 req->rq_rep_swab_mask = 0;
1087 * Used by ptlrpc client, to perform security transformation upon the reply
1088 * message of \a req. After return successfully, req->rq_repmsg points to
1089 * the reply message in clear text.
1091 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1094 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1096 LASSERT(req->rq_repbuf);
1097 LASSERT(req->rq_repdata == NULL);
1098 LASSERT(req->rq_repmsg == NULL);
1099 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1101 if (req->rq_reply_off == 0 &&
1102 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1103 CERROR("real reply with offset 0\n");
1107 if (req->rq_reply_off % 8 != 0) {
1108 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1112 req->rq_repdata = (struct lustre_msg *)
1113 (req->rq_repbuf + req->rq_reply_off);
1114 req->rq_repdata_len = req->rq_nob_received;
1116 return do_cli_unwrap_reply(req);
1120 * Used by ptlrpc client, to perform security transformation upon the early
1121 * reply message of \a req. We expect the rq_reply_off is 0, and
1122 * rq_nob_received is the early reply size.
1124 * Because the receive buffer might be still posted, the reply data might be
1125 * changed at any time, no matter we're holding rq_lock or not. For this reason
1126 * we allocate a separate ptlrpc_request and reply buffer for early reply
1129 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1130 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1131 * \a *req_ret to release it.
1132 * \retval -ev error number, and \a req_ret will not be set.
1134 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1135 struct ptlrpc_request **req_ret)
1137 struct ptlrpc_request *early_req;
1139 int early_bufsz, early_size;
1143 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1144 if (early_req == NULL)
1147 ptlrpc_cli_req_init(early_req);
1149 early_size = req->rq_nob_received;
1150 early_bufsz = size_roundup_power2(early_size);
1151 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1152 if (early_buf == NULL)
1153 GOTO(err_req, rc = -ENOMEM);
1155 /* sanity checkings and copy data out, do it inside spinlock */
1156 spin_lock(&req->rq_lock);
1158 if (req->rq_replied) {
1159 spin_unlock(&req->rq_lock);
1160 GOTO(err_buf, rc = -EALREADY);
1163 LASSERT(req->rq_repbuf);
1164 LASSERT(req->rq_repdata == NULL);
1165 LASSERT(req->rq_repmsg == NULL);
1167 if (req->rq_reply_off != 0) {
1168 CERROR("early reply with offset %u\n", req->rq_reply_off);
1169 spin_unlock(&req->rq_lock);
1170 GOTO(err_buf, rc = -EPROTO);
1173 if (req->rq_nob_received != early_size) {
1174 /* even another early arrived the size should be the same */
1175 CERROR("data size has changed from %u to %u\n",
1176 early_size, req->rq_nob_received);
1177 spin_unlock(&req->rq_lock);
1178 GOTO(err_buf, rc = -EINVAL);
1181 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1182 CERROR("early reply length %d too small\n",
1183 req->rq_nob_received);
1184 spin_unlock(&req->rq_lock);
1185 GOTO(err_buf, rc = -EALREADY);
1188 memcpy(early_buf, req->rq_repbuf, early_size);
1189 spin_unlock(&req->rq_lock);
1191 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1192 early_req->rq_flvr = req->rq_flvr;
1193 early_req->rq_repbuf = early_buf;
1194 early_req->rq_repbuf_len = early_bufsz;
1195 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1196 early_req->rq_repdata_len = early_size;
1197 early_req->rq_early = 1;
1198 early_req->rq_reqmsg = req->rq_reqmsg;
1200 rc = do_cli_unwrap_reply(early_req);
1202 DEBUG_REQ(D_ADAPTTO, early_req,
1203 "error %d unwrap early reply", rc);
1207 LASSERT(early_req->rq_repmsg);
1208 *req_ret = early_req;
1212 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1214 OBD_FREE_LARGE(early_buf, early_bufsz);
1216 ptlrpc_request_cache_free(early_req);
1221 * Used by ptlrpc client, to release a processed early reply \a early_req.
1223 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1225 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1227 LASSERT(early_req->rq_repbuf);
1228 LASSERT(early_req->rq_repdata);
1229 LASSERT(early_req->rq_repmsg);
1231 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1232 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1233 ptlrpc_request_cache_free(early_req);
1236 /**************************************************
1238 **************************************************/
1241 * "fixed" sec (e.g. null) use sec_id < 0
1243 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1245 int sptlrpc_get_next_secid(void)
1247 return atomic_inc_return(&sptlrpc_sec_id);
1249 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1251 /**************************************************
1252 * client side high-level security APIs *
1253 **************************************************/
1255 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1256 int grace, int force)
1258 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1260 LASSERT(policy->sp_cops);
1261 LASSERT(policy->sp_cops->flush_ctx_cache);
1263 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1266 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1268 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1270 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1271 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1272 LASSERT(policy->sp_cops->destroy_sec);
1274 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1276 policy->sp_cops->destroy_sec(sec);
1277 sptlrpc_policy_put(policy);
1280 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1282 sec_cop_destroy_sec(sec);
1284 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1286 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1288 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1290 if (sec->ps_policy->sp_cops->kill_sec) {
1291 sec->ps_policy->sp_cops->kill_sec(sec);
1293 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1297 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1300 atomic_inc(&sec->ps_refcount);
1304 EXPORT_SYMBOL(sptlrpc_sec_get);
1306 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1309 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1311 if (atomic_dec_and_test(&sec->ps_refcount)) {
1312 sptlrpc_gc_del_sec(sec);
1313 sec_cop_destroy_sec(sec);
1317 EXPORT_SYMBOL(sptlrpc_sec_put);
1320 * policy module is responsible for taking refrence of import
1323 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1324 struct ptlrpc_svc_ctx *svc_ctx,
1325 struct sptlrpc_flavor *sf,
1326 enum lustre_sec_part sp)
1328 struct ptlrpc_sec_policy *policy;
1329 struct ptlrpc_sec *sec;
1334 LASSERT(imp->imp_dlm_fake == 1);
1336 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1337 imp->imp_obd->obd_type->typ_name,
1338 imp->imp_obd->obd_name,
1339 sptlrpc_flavor2name(sf, str, sizeof(str)));
1341 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1342 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1344 LASSERT(imp->imp_dlm_fake == 0);
1346 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1347 imp->imp_obd->obd_type->typ_name,
1348 imp->imp_obd->obd_name,
1349 sptlrpc_flavor2name(sf, str, sizeof(str)));
1351 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1353 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1358 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1360 atomic_inc(&sec->ps_refcount);
1364 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1365 sptlrpc_gc_add_sec(sec);
1367 sptlrpc_policy_put(policy);
1373 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1375 struct ptlrpc_sec *sec;
1377 spin_lock(&imp->imp_lock);
1378 sec = sptlrpc_sec_get(imp->imp_sec);
1379 spin_unlock(&imp->imp_lock);
1383 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1385 static void sptlrpc_import_sec_install(struct obd_import *imp,
1386 struct ptlrpc_sec *sec)
1388 struct ptlrpc_sec *old_sec;
1390 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1392 spin_lock(&imp->imp_lock);
1393 old_sec = imp->imp_sec;
1395 spin_unlock(&imp->imp_lock);
1398 sptlrpc_sec_kill(old_sec);
1400 /* balance the ref taken by this import */
1401 sptlrpc_sec_put(old_sec);
1406 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1408 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1412 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1418 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1419 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1421 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1422 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1424 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1425 struct ptlrpc_svc_ctx *svc_ctx,
1426 struct sptlrpc_flavor *flvr)
1428 struct ptlrpc_connection *conn;
1429 struct sptlrpc_flavor sf;
1430 struct ptlrpc_sec *sec, *newsec;
1431 enum lustre_sec_part sp;
1441 conn = imp->imp_connection;
1443 if (svc_ctx == NULL) {
1444 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1446 * normal import, determine flavor from rule set, except
1447 * for mgc the flavor is predetermined.
1449 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1450 sf = cliobd->cl_flvr_mgc;
1452 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1454 &cliobd->cl_target_uuid,
1457 sp = imp->imp_obd->u.cli.cl_sp_me;
1459 /* reverse import, determine flavor from incoming reqeust */
1462 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1463 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1464 PTLRPC_SEC_FL_ROOTONLY;
1466 sp = sptlrpc_target_sec_part(imp->imp_obd);
1469 sec = sptlrpc_import_sec_ref(imp);
1473 if (flavor_equal(&sf, &sec->ps_flvr))
1476 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1477 imp->imp_obd->obd_name,
1478 obd_uuid2str(&conn->c_remote_uuid),
1479 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1480 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1481 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1482 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1483 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1484 imp->imp_obd->obd_name,
1485 obd_uuid2str(&conn->c_remote_uuid),
1486 LNET_NIDNET(conn->c_self),
1487 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1490 mutex_lock(&imp->imp_sec_mutex);
1492 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1494 sptlrpc_import_sec_install(imp, newsec);
1496 CERROR("import %s->%s: failed to create new sec\n",
1497 imp->imp_obd->obd_name,
1498 obd_uuid2str(&conn->c_remote_uuid));
1502 mutex_unlock(&imp->imp_sec_mutex);
1504 sptlrpc_sec_put(sec);
1508 void sptlrpc_import_sec_put(struct obd_import *imp)
1511 sptlrpc_sec_kill(imp->imp_sec);
1513 sptlrpc_sec_put(imp->imp_sec);
1514 imp->imp_sec = NULL;
1518 static void import_flush_ctx_common(struct obd_import *imp,
1519 uid_t uid, int grace, int force)
1521 struct ptlrpc_sec *sec;
1526 sec = sptlrpc_import_sec_ref(imp);
1530 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1531 sptlrpc_sec_put(sec);
1534 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1536 /* it's important to use grace mode, see explain in
1537 * sptlrpc_req_refresh_ctx() */
1538 import_flush_ctx_common(imp, 0, 1, 1);
1541 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1543 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1546 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1548 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1550 import_flush_ctx_common(imp, -1, 1, 1);
1552 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1555 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1556 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1558 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1560 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1561 struct ptlrpc_sec_policy *policy;
1565 LASSERT(ctx->cc_sec);
1566 LASSERT(ctx->cc_sec->ps_policy);
1567 LASSERT(req->rq_reqmsg == NULL);
1568 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1570 policy = ctx->cc_sec->ps_policy;
1571 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1573 LASSERT(req->rq_reqmsg);
1574 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1576 /* zeroing preallocated buffer */
1578 memset(req->rq_reqmsg, 0, msgsize);
1585 * Used by ptlrpc client to free request buffer of \a req. After this
1586 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1588 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1590 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1591 struct ptlrpc_sec_policy *policy;
1594 LASSERT(ctx->cc_sec);
1595 LASSERT(ctx->cc_sec->ps_policy);
1596 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1598 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1601 policy = ctx->cc_sec->ps_policy;
1602 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1603 req->rq_reqmsg = NULL;
1607 * NOTE caller must guarantee the buffer size is enough for the enlargement
1609 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1610 int segment, int newsize)
1613 int oldsize, oldmsg_size, movesize;
1615 LASSERT(segment < msg->lm_bufcount);
1616 LASSERT(msg->lm_buflens[segment] <= newsize);
1618 if (msg->lm_buflens[segment] == newsize)
1621 /* nothing to do if we are enlarging the last segment */
1622 if (segment == msg->lm_bufcount - 1) {
1623 msg->lm_buflens[segment] = newsize;
1627 oldsize = msg->lm_buflens[segment];
1629 src = lustre_msg_buf(msg, segment + 1, 0);
1630 msg->lm_buflens[segment] = newsize;
1631 dst = lustre_msg_buf(msg, segment + 1, 0);
1632 msg->lm_buflens[segment] = oldsize;
1634 /* move from segment + 1 to end segment */
1635 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1636 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1637 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1638 LASSERT(movesize >= 0);
1641 memmove(dst, src, movesize);
1643 /* note we don't clear the ares where old data live, not secret */
1645 /* finally set new segment size */
1646 msg->lm_buflens[segment] = newsize;
1648 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1651 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1652 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1653 * preserved after the enlargement. this must be called after original request
1654 * buffer being allocated.
1656 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1657 * so caller should refresh its local pointers if needed.
1659 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1660 const struct req_msg_field *field,
1663 struct req_capsule *pill = &req->rq_pill;
1664 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1665 struct ptlrpc_sec_cops *cops;
1666 struct lustre_msg *msg = req->rq_reqmsg;
1667 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1671 LASSERT(msg->lm_bufcount > segment);
1672 LASSERT(msg->lm_buflens[segment] <= newsize);
1674 if (msg->lm_buflens[segment] == newsize)
1677 cops = ctx->cc_sec->ps_policy->sp_cops;
1678 LASSERT(cops->enlarge_reqbuf);
1679 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1681 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1684 * Used by ptlrpc client to allocate reply buffer of \a req.
1686 * \note After this, req->rq_repmsg is still not accessible.
1688 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1690 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1691 struct ptlrpc_sec_policy *policy;
1695 LASSERT(ctx->cc_sec);
1696 LASSERT(ctx->cc_sec->ps_policy);
1701 policy = ctx->cc_sec->ps_policy;
1702 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1706 * Used by ptlrpc client to free reply buffer of \a req. After this
1707 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1709 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1711 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1712 struct ptlrpc_sec_policy *policy;
1716 LASSERT(ctx->cc_sec);
1717 LASSERT(ctx->cc_sec->ps_policy);
1718 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1720 if (req->rq_repbuf == NULL)
1722 LASSERT(req->rq_repbuf_len);
1724 policy = ctx->cc_sec->ps_policy;
1725 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1726 req->rq_repmsg = NULL;
1730 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1731 struct ptlrpc_cli_ctx *ctx)
1733 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1735 if (!policy->sp_cops->install_rctx)
1737 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1740 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1741 struct ptlrpc_svc_ctx *ctx)
1743 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1745 if (!policy->sp_sops->install_rctx)
1747 return policy->sp_sops->install_rctx(imp, ctx);
1750 /****************************************
1751 * server side security *
1752 ****************************************/
1754 static int flavor_allowed(struct sptlrpc_flavor *exp,
1755 struct ptlrpc_request *req)
1757 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1759 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1762 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1763 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1764 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1765 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1771 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1774 * Given an export \a exp, check whether the flavor of incoming \a req
1775 * is allowed by the export \a exp. Main logic is about taking care of
1776 * changing configurations. Return 0 means success.
1778 int sptlrpc_target_export_check(struct obd_export *exp,
1779 struct ptlrpc_request *req)
1781 struct sptlrpc_flavor flavor;
1786 /* client side export has no imp_reverse, skip
1787 * FIXME maybe we should check flavor this as well??? */
1788 if (exp->exp_imp_reverse == NULL)
1791 /* don't care about ctx fini rpc */
1792 if (req->rq_ctx_fini)
1795 spin_lock(&exp->exp_lock);
1797 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1798 * the first req with the new flavor, then treat it as current flavor,
1799 * adapt reverse sec according to it.
1800 * note the first rpc with new flavor might not be with root ctx, in
1801 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1802 if (unlikely(exp->exp_flvr_changed) &&
1803 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1804 /* make the new flavor as "current", and old ones as
1805 * about-to-expire */
1806 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1807 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1808 flavor = exp->exp_flvr_old[1];
1809 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1810 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1811 exp->exp_flvr_old[0] = exp->exp_flvr;
1812 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1813 EXP_FLVR_UPDATE_EXPIRE;
1814 exp->exp_flvr = flavor;
1816 /* flavor change finished */
1817 exp->exp_flvr_changed = 0;
1818 LASSERT(exp->exp_flvr_adapt == 1);
1820 /* if it's gss, we only interested in root ctx init */
1821 if (req->rq_auth_gss &&
1822 !(req->rq_ctx_init &&
1823 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1824 req->rq_auth_usr_ost))) {
1825 spin_unlock(&exp->exp_lock);
1826 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1827 req->rq_auth_gss, req->rq_ctx_init,
1828 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1829 req->rq_auth_usr_ost);
1833 exp->exp_flvr_adapt = 0;
1834 spin_unlock(&exp->exp_lock);
1836 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1837 req->rq_svc_ctx, &flavor);
1840 /* if it equals to the current flavor, we accept it, but need to
1841 * dealing with reverse sec/ctx */
1842 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1843 /* most cases should return here, we only interested in
1844 * gss root ctx init */
1845 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1846 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1847 !req->rq_auth_usr_ost)) {
1848 spin_unlock(&exp->exp_lock);
1852 /* if flavor just changed, we should not proceed, just leave
1853 * it and current flavor will be discovered and replaced
1854 * shortly, and let _this_ rpc pass through */
1855 if (exp->exp_flvr_changed) {
1856 LASSERT(exp->exp_flvr_adapt);
1857 spin_unlock(&exp->exp_lock);
1861 if (exp->exp_flvr_adapt) {
1862 exp->exp_flvr_adapt = 0;
1863 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1864 exp, exp->exp_flvr.sf_rpc,
1865 exp->exp_flvr_old[0].sf_rpc,
1866 exp->exp_flvr_old[1].sf_rpc);
1867 flavor = exp->exp_flvr;
1868 spin_unlock(&exp->exp_lock);
1870 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1874 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1875 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1876 exp->exp_flvr_old[0].sf_rpc,
1877 exp->exp_flvr_old[1].sf_rpc);
1878 spin_unlock(&exp->exp_lock);
1880 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1885 if (exp->exp_flvr_expire[0]) {
1886 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
1887 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1888 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
1889 exp->exp_flvr.sf_rpc,
1890 exp->exp_flvr_old[0].sf_rpc,
1891 exp->exp_flvr_old[1].sf_rpc,
1892 (s64)(exp->exp_flvr_expire[0] -
1893 ktime_get_real_seconds()));
1894 spin_unlock(&exp->exp_lock);
1898 CDEBUG(D_SEC, "mark middle expired\n");
1899 exp->exp_flvr_expire[0] = 0;
1901 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1902 exp->exp_flvr.sf_rpc,
1903 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1904 req->rq_flvr.sf_rpc);
1907 /* now it doesn't match the current flavor, the only chance we can
1908 * accept it is match the old flavors which is not expired. */
1909 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1910 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
1911 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1912 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
1914 exp->exp_flvr.sf_rpc,
1915 exp->exp_flvr_old[0].sf_rpc,
1916 exp->exp_flvr_old[1].sf_rpc,
1917 (s64)(exp->exp_flvr_expire[1] -
1918 ktime_get_real_seconds()));
1919 spin_unlock(&exp->exp_lock);
1923 CDEBUG(D_SEC, "mark oldest expired\n");
1924 exp->exp_flvr_expire[1] = 0;
1926 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1927 exp, exp->exp_flvr.sf_rpc,
1928 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1929 req->rq_flvr.sf_rpc);
1931 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1932 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1933 exp->exp_flvr_old[1].sf_rpc);
1936 spin_unlock(&exp->exp_lock);
1938 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
1939 exp, exp->exp_obd->obd_name,
1940 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1941 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1942 req->rq_flvr.sf_rpc,
1943 exp->exp_flvr.sf_rpc,
1944 exp->exp_flvr_old[0].sf_rpc,
1945 exp->exp_flvr_expire[0] ?
1946 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
1947 exp->exp_flvr_old[1].sf_rpc,
1948 exp->exp_flvr_expire[1] ?
1949 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
1952 EXPORT_SYMBOL(sptlrpc_target_export_check);
1954 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1955 struct sptlrpc_rule_set *rset)
1957 struct obd_export *exp;
1958 struct sptlrpc_flavor new_flvr;
1962 spin_lock(&obd->obd_dev_lock);
1964 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1965 if (exp->exp_connection == NULL)
1968 /* note if this export had just been updated flavor
1969 * (exp_flvr_changed == 1), this will override the
1971 spin_lock(&exp->exp_lock);
1972 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1973 exp->exp_connection->c_peer.nid,
1975 if (exp->exp_flvr_changed ||
1976 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1977 exp->exp_flvr_old[1] = new_flvr;
1978 exp->exp_flvr_expire[1] = 0;
1979 exp->exp_flvr_changed = 1;
1980 exp->exp_flvr_adapt = 1;
1982 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1983 exp, sptlrpc_part2name(exp->exp_sp_peer),
1984 exp->exp_flvr.sf_rpc,
1985 exp->exp_flvr_old[1].sf_rpc);
1987 spin_unlock(&exp->exp_lock);
1990 spin_unlock(&obd->obd_dev_lock);
1992 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1994 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1996 /* peer's claim is unreliable unless gss is being used */
1997 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2000 switch (req->rq_sp_from) {
2002 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2003 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2004 svc_rc = SECSVC_DROP;
2008 if (!req->rq_auth_usr_mdt) {
2009 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2010 svc_rc = SECSVC_DROP;
2014 if (!req->rq_auth_usr_ost) {
2015 DEBUG_REQ(D_ERROR, req, "faked source OST");
2016 svc_rc = SECSVC_DROP;
2021 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2022 !req->rq_auth_usr_ost) {
2023 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2024 svc_rc = SECSVC_DROP;
2029 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2030 svc_rc = SECSVC_DROP;
2037 * Used by ptlrpc server, to perform transformation upon request message of
2038 * incoming \a req. This must be the first thing to do with an incoming
2039 * request in ptlrpc layer.
2041 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2042 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2043 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2044 * reply message has been prepared.
2045 * \retval SECSVC_DROP failed, this request should be dropped.
2047 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2049 struct ptlrpc_sec_policy *policy;
2050 struct lustre_msg *msg = req->rq_reqbuf;
2055 LASSERT(req->rq_reqmsg == NULL);
2056 LASSERT(req->rq_repmsg == NULL);
2057 LASSERT(req->rq_svc_ctx == NULL);
2059 req->rq_req_swab_mask = 0;
2061 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2064 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2068 CERROR("error unpacking request from %s x%llu\n",
2069 libcfs_id2str(req->rq_peer), req->rq_xid);
2070 RETURN(SECSVC_DROP);
2073 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2074 req->rq_sp_from = LUSTRE_SP_ANY;
2075 req->rq_auth_uid = -1; /* set to INVALID_UID */
2076 req->rq_auth_mapped_uid = -1;
2078 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2080 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2081 RETURN(SECSVC_DROP);
2084 LASSERT(policy->sp_sops->accept);
2085 rc = policy->sp_sops->accept(req);
2086 sptlrpc_policy_put(policy);
2087 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2088 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2091 * if it's not null flavor (which means embedded packing msg),
2092 * reset the swab mask for the comming inner msg unpacking.
2094 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2095 req->rq_req_swab_mask = 0;
2097 /* sanity check for the request source */
2098 rc = sptlrpc_svc_check_from(req, rc);
2103 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2104 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2105 * a buffer of \a msglen size.
2107 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2109 struct ptlrpc_sec_policy *policy;
2110 struct ptlrpc_reply_state *rs;
2114 LASSERT(req->rq_svc_ctx);
2115 LASSERT(req->rq_svc_ctx->sc_policy);
2117 policy = req->rq_svc_ctx->sc_policy;
2118 LASSERT(policy->sp_sops->alloc_rs);
2120 rc = policy->sp_sops->alloc_rs(req, msglen);
2121 if (unlikely(rc == -ENOMEM)) {
2122 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2123 if (svcpt->scp_service->srv_max_reply_size <
2124 msglen + sizeof(struct ptlrpc_reply_state)) {
2125 /* Just return failure if the size is too big */
2126 CERROR("size of message is too big (%zd), %d allowed\n",
2127 msglen + sizeof(struct ptlrpc_reply_state),
2128 svcpt->scp_service->srv_max_reply_size);
2132 /* failed alloc, try emergency pool */
2133 rs = lustre_get_emerg_rs(svcpt);
2137 req->rq_reply_state = rs;
2138 rc = policy->sp_sops->alloc_rs(req, msglen);
2140 lustre_put_emerg_rs(rs);
2141 req->rq_reply_state = NULL;
2146 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2152 * Used by ptlrpc server, to perform transformation upon reply message.
2154 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2155 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2157 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2159 struct ptlrpc_sec_policy *policy;
2163 LASSERT(req->rq_svc_ctx);
2164 LASSERT(req->rq_svc_ctx->sc_policy);
2166 policy = req->rq_svc_ctx->sc_policy;
2167 LASSERT(policy->sp_sops->authorize);
2169 rc = policy->sp_sops->authorize(req);
2170 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2176 * Used by ptlrpc server, to free reply_state.
2178 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2180 struct ptlrpc_sec_policy *policy;
2181 unsigned int prealloc;
2184 LASSERT(rs->rs_svc_ctx);
2185 LASSERT(rs->rs_svc_ctx->sc_policy);
2187 policy = rs->rs_svc_ctx->sc_policy;
2188 LASSERT(policy->sp_sops->free_rs);
2190 prealloc = rs->rs_prealloc;
2191 policy->sp_sops->free_rs(rs);
2194 lustre_put_emerg_rs(rs);
2198 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2200 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2203 atomic_inc(&ctx->sc_refcount);
2206 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2208 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2213 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2214 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2215 if (ctx->sc_policy->sp_sops->free_ctx)
2216 ctx->sc_policy->sp_sops->free_ctx(ctx);
2218 req->rq_svc_ctx = NULL;
2221 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2223 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2228 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2229 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2230 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2232 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2234 /****************************************
2236 ****************************************/
2239 * Perform transformation upon bulk data pointed by \a desc. This is called
2240 * before transforming the request message.
2242 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2243 struct ptlrpc_bulk_desc *desc)
2245 struct ptlrpc_cli_ctx *ctx;
2247 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2249 if (!req->rq_pack_bulk)
2252 ctx = req->rq_cli_ctx;
2253 if (ctx->cc_ops->wrap_bulk)
2254 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2257 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2260 * This is called after unwrap the reply message.
2261 * return nob of actual plain text size received, or error code.
2263 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2264 struct ptlrpc_bulk_desc *desc,
2267 struct ptlrpc_cli_ctx *ctx;
2270 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2272 if (!req->rq_pack_bulk)
2273 return desc->bd_nob_transferred;
2275 ctx = req->rq_cli_ctx;
2276 if (ctx->cc_ops->unwrap_bulk) {
2277 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2281 return desc->bd_nob_transferred;
2283 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2286 * This is called after unwrap the reply message.
2287 * return 0 for success or error code.
2289 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2290 struct ptlrpc_bulk_desc *desc)
2292 struct ptlrpc_cli_ctx *ctx;
2295 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2297 if (!req->rq_pack_bulk)
2300 ctx = req->rq_cli_ctx;
2301 if (ctx->cc_ops->unwrap_bulk) {
2302 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2308 * if everything is going right, nob should equals to nob_transferred.
2309 * in case of privacy mode, nob_transferred needs to be adjusted.
2311 if (desc->bd_nob != desc->bd_nob_transferred) {
2312 CERROR("nob %d doesn't match transferred nob %d\n",
2313 desc->bd_nob, desc->bd_nob_transferred);
2319 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2321 #ifdef HAVE_SERVER_SUPPORT
2323 * Performe transformation upon outgoing bulk read.
2325 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2326 struct ptlrpc_bulk_desc *desc)
2328 struct ptlrpc_svc_ctx *ctx;
2330 LASSERT(req->rq_bulk_read);
2332 if (!req->rq_pack_bulk)
2335 ctx = req->rq_svc_ctx;
2336 if (ctx->sc_policy->sp_sops->wrap_bulk)
2337 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2341 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2344 * Performe transformation upon incoming bulk write.
2346 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2347 struct ptlrpc_bulk_desc *desc)
2349 struct ptlrpc_svc_ctx *ctx;
2352 LASSERT(req->rq_bulk_write);
2355 * if it's in privacy mode, transferred should >= expected; otherwise
2356 * transferred should == expected.
2358 if (desc->bd_nob_transferred < desc->bd_nob ||
2359 (desc->bd_nob_transferred > desc->bd_nob &&
2360 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2361 SPTLRPC_BULK_SVC_PRIV)) {
2362 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2363 desc->bd_nob_transferred, desc->bd_nob);
2367 if (!req->rq_pack_bulk)
2370 ctx = req->rq_svc_ctx;
2371 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2372 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2374 CERROR("error unwrap bulk: %d\n", rc);
2377 /* return 0 to allow reply be sent */
2380 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2383 * Prepare buffers for incoming bulk write.
2385 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2386 struct ptlrpc_bulk_desc *desc)
2388 struct ptlrpc_svc_ctx *ctx;
2390 LASSERT(req->rq_bulk_write);
2392 if (!req->rq_pack_bulk)
2395 ctx = req->rq_svc_ctx;
2396 if (ctx->sc_policy->sp_sops->prep_bulk)
2397 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2401 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2403 #endif /* HAVE_SERVER_SUPPORT */
2405 /****************************************
2406 * user descriptor helpers *
2407 ****************************************/
2409 int sptlrpc_current_user_desc_size(void)
2413 ngroups = current_ngroups;
2415 if (ngroups > LUSTRE_MAX_GROUPS)
2416 ngroups = LUSTRE_MAX_GROUPS;
2417 return sptlrpc_user_desc_size(ngroups);
2419 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2421 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2423 struct ptlrpc_user_desc *pud;
2425 pud = lustre_msg_buf(msg, offset, 0);
2427 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2428 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2429 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2430 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2431 pud->pud_cap = cfs_curproc_cap_pack();
2432 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2435 if (pud->pud_ngroups > current_ngroups)
2436 pud->pud_ngroups = current_ngroups;
2437 #ifdef HAVE_GROUP_INFO_GID
2438 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2439 pud->pud_ngroups * sizeof(__u32));
2440 #else /* !HAVE_GROUP_INFO_GID */
2441 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2442 pud->pud_ngroups * sizeof(__u32));
2443 #endif /* HAVE_GROUP_INFO_GID */
2444 task_unlock(current);
2448 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2450 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2452 struct ptlrpc_user_desc *pud;
2455 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2460 __swab32s(&pud->pud_uid);
2461 __swab32s(&pud->pud_gid);
2462 __swab32s(&pud->pud_fsuid);
2463 __swab32s(&pud->pud_fsgid);
2464 __swab32s(&pud->pud_cap);
2465 __swab32s(&pud->pud_ngroups);
2468 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2469 CERROR("%u groups is too large\n", pud->pud_ngroups);
2473 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2474 msg->lm_buflens[offset]) {
2475 CERROR("%u groups are claimed but bufsize only %u\n",
2476 pud->pud_ngroups, msg->lm_buflens[offset]);
2481 for (i = 0; i < pud->pud_ngroups; i++)
2482 __swab32s(&pud->pud_groups[i]);
2487 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2489 /****************************************
2491 ****************************************/
2493 const char * sec2target_str(struct ptlrpc_sec *sec)
2495 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2497 if (sec_is_reverse(sec))
2499 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2501 EXPORT_SYMBOL(sec2target_str);
2504 * return true if the bulk data is protected
2506 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2508 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2509 case SPTLRPC_BULK_SVC_INTG:
2510 case SPTLRPC_BULK_SVC_PRIV:
2516 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2518 /****************************************
2519 * crypto API helper/alloc blkciper *
2520 ****************************************/
2522 /****************************************
2523 * initialize/finalize *
2524 ****************************************/
2526 int sptlrpc_init(void)
2530 rwlock_init(&policy_lock);
2532 rc = sptlrpc_gc_init();
2536 rc = sptlrpc_conf_init();
2540 rc = sptlrpc_enc_pool_init();
2544 rc = sptlrpc_null_init();
2548 rc = sptlrpc_plain_init();
2552 rc = sptlrpc_lproc_init();
2559 sptlrpc_plain_fini();
2561 sptlrpc_null_fini();
2563 sptlrpc_enc_pool_fini();
2565 sptlrpc_conf_fini();
2572 void sptlrpc_fini(void)
2574 sptlrpc_lproc_fini();
2575 sptlrpc_plain_fini();
2576 sptlrpc_null_fini();
2577 sptlrpc_enc_pool_fini();
2578 sptlrpc_conf_fini();