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/
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);
406 CERROR("import %p (%s) with no sec\n",
407 imp, ptlrpc_import_state_name(imp->imp_state));
411 if (unlikely((*sec)->ps_dying)) {
412 CERROR("attempt to use dying sec %p\n", sec);
413 sptlrpc_sec_put(*sec);
421 * Given a \a req, find or allocate an appropriate context for it.
422 * \pre req->rq_cli_ctx == NULL.
424 * \retval 0 succeed, and req->rq_cli_ctx is set.
425 * \retval -ev error number, and req->rq_cli_ctx == NULL.
427 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
429 struct obd_import *imp = req->rq_import;
430 struct ptlrpc_sec *sec;
434 LASSERT(!req->rq_cli_ctx);
437 rc = import_sec_validate_get(imp, &sec);
441 req->rq_cli_ctx = get_my_ctx(sec);
443 sptlrpc_sec_put(sec);
445 if (!req->rq_cli_ctx) {
446 CERROR("req %p: fail to get context\n", req);
447 RETURN(-ECONNREFUSED);
454 * Drop the context for \a req.
455 * \pre req->rq_cli_ctx != NULL.
456 * \post req->rq_cli_ctx == NULL.
458 * If \a sync == 0, this function should return quickly without sleep;
459 * otherwise it might trigger and wait for the whole process of sending
460 * an context-destroying rpc to server.
462 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
467 LASSERT(req->rq_cli_ctx);
469 /* request might be asked to release earlier while still
470 * in the context waiting list.
472 if (!list_empty(&req->rq_ctx_chain)) {
473 spin_lock(&req->rq_cli_ctx->cc_lock);
474 list_del_init(&req->rq_ctx_chain);
475 spin_unlock(&req->rq_cli_ctx->cc_lock);
478 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
479 req->rq_cli_ctx = NULL;
484 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
485 struct ptlrpc_cli_ctx *oldctx,
486 struct ptlrpc_cli_ctx *newctx)
488 struct sptlrpc_flavor old_flvr;
489 char *reqmsg = NULL; /* to workaround old gcc */
493 LASSERT(req->rq_reqmsg);
494 LASSERT(req->rq_reqlen);
495 LASSERT(req->rq_replen);
497 CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
498 "switch sec %p(%s) -> %p(%s)\n", req,
499 oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
500 newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
501 oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
502 newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
505 old_flvr = req->rq_flvr;
507 /* save request message */
508 reqmsg_size = req->rq_reqlen;
509 if (reqmsg_size != 0) {
510 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
513 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
516 /* release old req/rep buf */
517 req->rq_cli_ctx = oldctx;
518 sptlrpc_cli_free_reqbuf(req);
519 sptlrpc_cli_free_repbuf(req);
520 req->rq_cli_ctx = newctx;
522 /* recalculate the flavor */
523 sptlrpc_req_set_flavor(req, 0);
525 /* alloc new request buffer
526 * we don't need to alloc reply buffer here, leave it to the
527 * rest procedure of ptlrpc */
528 if (reqmsg_size != 0) {
529 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
531 LASSERT(req->rq_reqmsg);
532 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
534 CWARN("failed to alloc reqbuf: %d\n", rc);
535 req->rq_flvr = old_flvr;
538 OBD_FREE_LARGE(reqmsg, reqmsg_size);
544 * If current context of \a req is dead somehow, e.g. we just switched flavor
545 * thus marked original contexts dead, we'll find a new context for it. if
546 * no switch is needed, \a req will end up with the same context.
548 * \note a request must have a context, to keep other parts of code happy.
549 * In any case of failure during the switching, we must restore the old one.
551 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
553 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
554 struct ptlrpc_cli_ctx *newctx;
560 sptlrpc_cli_ctx_get(oldctx);
561 sptlrpc_req_put_ctx(req, 0);
563 rc = sptlrpc_req_get_ctx(req);
565 LASSERT(!req->rq_cli_ctx);
567 /* restore old ctx */
568 req->rq_cli_ctx = oldctx;
572 newctx = req->rq_cli_ctx;
575 if (unlikely(newctx == oldctx &&
576 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
578 * still get the old dead ctx, usually means system too busy
581 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
582 newctx, newctx->cc_flags);
584 set_current_state(TASK_INTERRUPTIBLE);
585 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
586 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
589 * new ctx not up to date yet
592 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
593 newctx, newctx->cc_flags);
596 * it's possible newctx == oldctx if we're switching
597 * subflavor with the same sec.
599 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
601 /* restore old ctx */
602 sptlrpc_req_put_ctx(req, 0);
603 req->rq_cli_ctx = oldctx;
607 LASSERT(req->rq_cli_ctx == newctx);
610 sptlrpc_cli_ctx_put(oldctx, 1);
613 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
616 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
618 if (cli_ctx_is_refreshed(ctx))
624 int ctx_refresh_timeout(void *data)
626 struct ptlrpc_request *req = data;
629 /* conn_cnt is needed in expire_one_request */
630 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
632 rc = ptlrpc_expire_one_request(req, 1);
633 /* if we started recovery, we should mark this ctx dead; otherwise
634 * in case of lgssd died nobody would retire this ctx, following
635 * connecting will still find the same ctx thus cause deadlock.
636 * there's an assumption that expire time of the request should be
637 * later than the context refresh expire time.
640 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
645 void ctx_refresh_interrupt(void *data)
647 struct ptlrpc_request *req = data;
649 spin_lock(&req->rq_lock);
651 spin_unlock(&req->rq_lock);
655 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
657 spin_lock(&ctx->cc_lock);
658 if (!list_empty(&req->rq_ctx_chain))
659 list_del_init(&req->rq_ctx_chain);
660 spin_unlock(&ctx->cc_lock);
664 * To refresh the context of \req, if it's not up-to-date.
667 * - = 0: wait until success or fatal error occur
668 * - > 0: timeout value (in seconds)
670 * The status of the context could be subject to be changed by other threads
671 * at any time. We allow this race, but once we return with 0, the caller will
672 * suppose it's uptodated and keep using it until the owning rpc is done.
674 * \retval 0 only if the context is uptodated.
675 * \retval -ev error number.
677 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
679 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
680 struct ptlrpc_sec *sec;
681 struct l_wait_info lwi;
687 if (req->rq_ctx_init || req->rq_ctx_fini)
691 * during the process a request's context might change type even
692 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
696 rc = import_sec_validate_get(req->rq_import, &sec);
700 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
701 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
702 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
703 req_off_ctx_list(req, ctx);
704 sptlrpc_req_replace_dead_ctx(req);
705 ctx = req->rq_cli_ctx;
707 sptlrpc_sec_put(sec);
709 if (cli_ctx_is_eternal(ctx))
712 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
713 LASSERT(ctx->cc_ops->refresh);
714 ctx->cc_ops->refresh(ctx);
716 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
718 LASSERT(ctx->cc_ops->validate);
719 if (ctx->cc_ops->validate(ctx) == 0) {
720 req_off_ctx_list(req, ctx);
724 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
725 spin_lock(&req->rq_lock);
727 spin_unlock(&req->rq_lock);
728 req_off_ctx_list(req, ctx);
733 * There's a subtle issue for resending RPCs, suppose following
735 * 1. the request was sent to server.
736 * 2. recovery was kicked start, after finished the request was
738 * 3. resend the request.
739 * 4. old reply from server received, we accept and verify the reply.
740 * this has to be success, otherwise the error will be aware
742 * 5. new reply from server received, dropped by LNet.
744 * Note the xid of old & new request is the same. We can't simply
745 * change xid for the resent request because the server replies on
746 * it for reply reconstruction.
748 * Commonly the original context should be uptodate because we
749 * have an expiry nice time; server will keep its context because
750 * we at least hold a ref of old context which prevent context
751 * from destroying RPC being sent. So server still can accept the
752 * request and finish the RPC. But if that's not the case:
753 * 1. If server side context has been trimmed, a NO_CONTEXT will
754 * be returned, gss_cli_ctx_verify/unseal will switch to new
756 * 2. Current context never be refreshed, then we are fine: we
757 * never really send request with old context before.
759 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
760 unlikely(req->rq_reqmsg) &&
761 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
762 req_off_ctx_list(req, ctx);
766 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
767 req_off_ctx_list(req, ctx);
769 * don't switch ctx if import was deactivated
771 if (req->rq_import->imp_deactive) {
772 spin_lock(&req->rq_lock);
774 spin_unlock(&req->rq_lock);
778 rc = sptlrpc_req_replace_dead_ctx(req);
780 LASSERT(ctx == req->rq_cli_ctx);
781 CERROR("req %p: failed to replace dead ctx %p: %d\n",
783 spin_lock(&req->rq_lock);
785 spin_unlock(&req->rq_lock);
789 ctx = req->rq_cli_ctx;
794 * Now we're sure this context is during upcall, add myself into
797 spin_lock(&ctx->cc_lock);
798 if (list_empty(&req->rq_ctx_chain))
799 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
800 spin_unlock(&ctx->cc_lock);
803 RETURN(-EWOULDBLOCK);
805 /* Clear any flags that may be present from previous sends */
806 LASSERT(req->rq_receiving_reply == 0);
807 spin_lock(&req->rq_lock);
809 req->rq_timedout = 0;
812 spin_unlock(&req->rq_lock);
814 lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
816 ctx_refresh_interrupt, req);
817 rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
820 * following cases could lead us here:
821 * - successfully refreshed;
823 * - timedout, and we don't want recover from the failure;
824 * - timedout, and waked up upon recovery finished;
825 * - someone else mark this ctx dead by force;
826 * - someone invalidate the req and call ptlrpc_client_wake_req(),
827 * e.g. ptlrpc_abort_inflight();
829 if (!cli_ctx_is_refreshed(ctx)) {
830 /* timed out or interruptted */
831 req_off_ctx_list(req, ctx);
841 * Initialize flavor settings for \a req, according to \a opcode.
843 * \note this could be called in two situations:
844 * - new request from ptlrpc_pre_req(), with proper @opcode
845 * - old request which changed ctx in the middle, with @opcode == 0
847 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
849 struct ptlrpc_sec *sec;
851 LASSERT(req->rq_import);
852 LASSERT(req->rq_cli_ctx);
853 LASSERT(req->rq_cli_ctx->cc_sec);
854 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
856 /* special security flags according to opcode */
860 case MGS_CONFIG_READ:
862 req->rq_bulk_read = 1;
866 req->rq_bulk_write = 1;
869 req->rq_ctx_init = 1;
872 req->rq_ctx_fini = 1;
875 /* init/fini rpc won't be resend, so can't be here */
876 LASSERT(req->rq_ctx_init == 0);
877 LASSERT(req->rq_ctx_fini == 0);
879 /* cleanup flags, which should be recalculated */
880 req->rq_pack_udesc = 0;
881 req->rq_pack_bulk = 0;
885 sec = req->rq_cli_ctx->cc_sec;
887 spin_lock(&sec->ps_lock);
888 req->rq_flvr = sec->ps_flvr;
889 spin_unlock(&sec->ps_lock);
891 /* force SVC_NULL for context initiation rpc, SVC_INTG for context
893 if (unlikely(req->rq_ctx_init))
894 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
895 else if (unlikely(req->rq_ctx_fini))
896 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
898 /* user descriptor flag, null security can't do it anyway */
899 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
900 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
901 req->rq_pack_udesc = 1;
903 /* bulk security flag */
904 if ((req->rq_bulk_read || req->rq_bulk_write) &&
905 sptlrpc_flavor_has_bulk(&req->rq_flvr))
906 req->rq_pack_bulk = 1;
909 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
911 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
914 LASSERT(req->rq_clrbuf);
915 if (req->rq_pool || !req->rq_reqbuf)
918 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
919 req->rq_reqbuf = NULL;
920 req->rq_reqbuf_len = 0;
924 * Given an import \a imp, check whether current user has a valid context
925 * or not. We may create a new context and try to refresh it, and try
926 * repeatedly try in case of non-fatal errors. Return 0 means success.
928 int sptlrpc_import_check_ctx(struct obd_import *imp)
930 struct ptlrpc_sec *sec;
931 struct ptlrpc_cli_ctx *ctx;
932 struct ptlrpc_request *req = NULL;
938 sec = sptlrpc_import_sec_ref(imp);
939 ctx = get_my_ctx(sec);
940 sptlrpc_sec_put(sec);
945 if (cli_ctx_is_eternal(ctx) ||
946 ctx->cc_ops->validate(ctx) == 0) {
947 sptlrpc_cli_ctx_put(ctx, 1);
951 if (cli_ctx_is_error(ctx)) {
952 sptlrpc_cli_ctx_put(ctx, 1);
956 req = ptlrpc_request_cache_alloc(GFP_NOFS);
960 ptlrpc_cli_req_init(req);
961 atomic_set(&req->rq_refcount, 10000);
963 req->rq_import = imp;
964 req->rq_flvr = sec->ps_flvr;
965 req->rq_cli_ctx = ctx;
967 rc = sptlrpc_req_refresh_ctx(req, 0);
968 LASSERT(list_empty(&req->rq_ctx_chain));
969 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
970 ptlrpc_request_cache_free(req);
976 * Used by ptlrpc client, to perform the pre-defined security transformation
977 * upon the request message of \a req. After this function called,
978 * req->rq_reqmsg is still accessible as clear text.
980 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
982 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
987 LASSERT(ctx->cc_sec);
988 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
990 /* we wrap bulk request here because now we can be sure
991 * the context is uptodate.
994 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
999 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1000 case SPTLRPC_SVC_NULL:
1001 case SPTLRPC_SVC_AUTH:
1002 case SPTLRPC_SVC_INTG:
1003 LASSERT(ctx->cc_ops->sign);
1004 rc = ctx->cc_ops->sign(ctx, req);
1006 case SPTLRPC_SVC_PRIV:
1007 LASSERT(ctx->cc_ops->seal);
1008 rc = ctx->cc_ops->seal(ctx, req);
1015 LASSERT(req->rq_reqdata_len);
1016 LASSERT(req->rq_reqdata_len % 8 == 0);
1017 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1023 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1025 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1030 LASSERT(ctx->cc_sec);
1031 LASSERT(req->rq_repbuf);
1032 LASSERT(req->rq_repdata);
1033 LASSERT(req->rq_repmsg == NULL);
1035 req->rq_rep_swab_mask = 0;
1037 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1040 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1044 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1048 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1049 CERROR("replied data length %d too small\n",
1050 req->rq_repdata_len);
1054 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1055 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1056 CERROR("reply policy %u doesn't match request policy %u\n",
1057 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1058 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1062 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1063 case SPTLRPC_SVC_NULL:
1064 case SPTLRPC_SVC_AUTH:
1065 case SPTLRPC_SVC_INTG:
1066 LASSERT(ctx->cc_ops->verify);
1067 rc = ctx->cc_ops->verify(ctx, req);
1069 case SPTLRPC_SVC_PRIV:
1070 LASSERT(ctx->cc_ops->unseal);
1071 rc = ctx->cc_ops->unseal(ctx, req);
1076 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1078 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1080 req->rq_rep_swab_mask = 0;
1085 * Used by ptlrpc client, to perform security transformation upon the reply
1086 * message of \a req. After return successfully, req->rq_repmsg points to
1087 * the reply message in clear text.
1089 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1092 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1094 LASSERT(req->rq_repbuf);
1095 LASSERT(req->rq_repdata == NULL);
1096 LASSERT(req->rq_repmsg == NULL);
1097 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1099 if (req->rq_reply_off == 0 &&
1100 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1101 CERROR("real reply with offset 0\n");
1105 if (req->rq_reply_off % 8 != 0) {
1106 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1110 req->rq_repdata = (struct lustre_msg *)
1111 (req->rq_repbuf + req->rq_reply_off);
1112 req->rq_repdata_len = req->rq_nob_received;
1114 return do_cli_unwrap_reply(req);
1118 * Used by ptlrpc client, to perform security transformation upon the early
1119 * reply message of \a req. We expect the rq_reply_off is 0, and
1120 * rq_nob_received is the early reply size.
1122 * Because the receive buffer might be still posted, the reply data might be
1123 * changed at any time, no matter we're holding rq_lock or not. For this reason
1124 * we allocate a separate ptlrpc_request and reply buffer for early reply
1127 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1128 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1129 * \a *req_ret to release it.
1130 * \retval -ev error number, and \a req_ret will not be set.
1132 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1133 struct ptlrpc_request **req_ret)
1135 struct ptlrpc_request *early_req;
1137 int early_bufsz, early_size;
1141 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1142 if (early_req == NULL)
1145 ptlrpc_cli_req_init(early_req);
1147 early_size = req->rq_nob_received;
1148 early_bufsz = size_roundup_power2(early_size);
1149 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1150 if (early_buf == NULL)
1151 GOTO(err_req, rc = -ENOMEM);
1153 /* sanity checkings and copy data out, do it inside spinlock */
1154 spin_lock(&req->rq_lock);
1156 if (req->rq_replied) {
1157 spin_unlock(&req->rq_lock);
1158 GOTO(err_buf, rc = -EALREADY);
1161 LASSERT(req->rq_repbuf);
1162 LASSERT(req->rq_repdata == NULL);
1163 LASSERT(req->rq_repmsg == NULL);
1165 if (req->rq_reply_off != 0) {
1166 CERROR("early reply with offset %u\n", req->rq_reply_off);
1167 spin_unlock(&req->rq_lock);
1168 GOTO(err_buf, rc = -EPROTO);
1171 if (req->rq_nob_received != early_size) {
1172 /* even another early arrived the size should be the same */
1173 CERROR("data size has changed from %u to %u\n",
1174 early_size, req->rq_nob_received);
1175 spin_unlock(&req->rq_lock);
1176 GOTO(err_buf, rc = -EINVAL);
1179 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1180 CERROR("early reply length %d too small\n",
1181 req->rq_nob_received);
1182 spin_unlock(&req->rq_lock);
1183 GOTO(err_buf, rc = -EALREADY);
1186 memcpy(early_buf, req->rq_repbuf, early_size);
1187 spin_unlock(&req->rq_lock);
1189 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1190 early_req->rq_flvr = req->rq_flvr;
1191 early_req->rq_repbuf = early_buf;
1192 early_req->rq_repbuf_len = early_bufsz;
1193 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1194 early_req->rq_repdata_len = early_size;
1195 early_req->rq_early = 1;
1196 early_req->rq_reqmsg = req->rq_reqmsg;
1198 rc = do_cli_unwrap_reply(early_req);
1200 DEBUG_REQ(D_ADAPTTO, early_req,
1201 "error %d unwrap early reply", rc);
1205 LASSERT(early_req->rq_repmsg);
1206 *req_ret = early_req;
1210 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1212 OBD_FREE_LARGE(early_buf, early_bufsz);
1214 ptlrpc_request_cache_free(early_req);
1219 * Used by ptlrpc client, to release a processed early reply \a early_req.
1221 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1223 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1225 LASSERT(early_req->rq_repbuf);
1226 LASSERT(early_req->rq_repdata);
1227 LASSERT(early_req->rq_repmsg);
1229 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1230 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1231 ptlrpc_request_cache_free(early_req);
1234 /**************************************************
1236 **************************************************/
1239 * "fixed" sec (e.g. null) use sec_id < 0
1241 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1243 int sptlrpc_get_next_secid(void)
1245 return atomic_inc_return(&sptlrpc_sec_id);
1247 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1249 /**************************************************
1250 * client side high-level security APIs *
1251 **************************************************/
1253 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1254 int grace, int force)
1256 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1258 LASSERT(policy->sp_cops);
1259 LASSERT(policy->sp_cops->flush_ctx_cache);
1261 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1264 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1266 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1268 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1269 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1270 LASSERT(policy->sp_cops->destroy_sec);
1272 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1274 policy->sp_cops->destroy_sec(sec);
1275 sptlrpc_policy_put(policy);
1278 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1280 sec_cop_destroy_sec(sec);
1282 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1284 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1286 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1288 if (sec->ps_policy->sp_cops->kill_sec) {
1289 sec->ps_policy->sp_cops->kill_sec(sec);
1291 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1295 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1298 atomic_inc(&sec->ps_refcount);
1302 EXPORT_SYMBOL(sptlrpc_sec_get);
1304 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1307 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1309 if (atomic_dec_and_test(&sec->ps_refcount)) {
1310 sptlrpc_gc_del_sec(sec);
1311 sec_cop_destroy_sec(sec);
1315 EXPORT_SYMBOL(sptlrpc_sec_put);
1318 * policy module is responsible for taking refrence of import
1321 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1322 struct ptlrpc_svc_ctx *svc_ctx,
1323 struct sptlrpc_flavor *sf,
1324 enum lustre_sec_part sp)
1326 struct ptlrpc_sec_policy *policy;
1327 struct ptlrpc_sec *sec;
1332 LASSERT(imp->imp_dlm_fake == 1);
1334 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1335 imp->imp_obd->obd_type->typ_name,
1336 imp->imp_obd->obd_name,
1337 sptlrpc_flavor2name(sf, str, sizeof(str)));
1339 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1340 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1342 LASSERT(imp->imp_dlm_fake == 0);
1344 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1345 imp->imp_obd->obd_type->typ_name,
1346 imp->imp_obd->obd_name,
1347 sptlrpc_flavor2name(sf, str, sizeof(str)));
1349 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1351 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1356 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1358 atomic_inc(&sec->ps_refcount);
1362 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1363 sptlrpc_gc_add_sec(sec);
1365 sptlrpc_policy_put(policy);
1371 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1373 struct ptlrpc_sec *sec;
1375 spin_lock(&imp->imp_lock);
1376 sec = sptlrpc_sec_get(imp->imp_sec);
1377 spin_unlock(&imp->imp_lock);
1381 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1383 static void sptlrpc_import_sec_install(struct obd_import *imp,
1384 struct ptlrpc_sec *sec)
1386 struct ptlrpc_sec *old_sec;
1388 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1390 spin_lock(&imp->imp_lock);
1391 old_sec = imp->imp_sec;
1393 spin_unlock(&imp->imp_lock);
1396 sptlrpc_sec_kill(old_sec);
1398 /* balance the ref taken by this import */
1399 sptlrpc_sec_put(old_sec);
1404 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1406 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1410 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1416 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1417 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1419 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1420 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1422 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1423 struct ptlrpc_svc_ctx *svc_ctx,
1424 struct sptlrpc_flavor *flvr)
1426 struct ptlrpc_connection *conn;
1427 struct sptlrpc_flavor sf;
1428 struct ptlrpc_sec *sec, *newsec;
1429 enum lustre_sec_part sp;
1439 conn = imp->imp_connection;
1441 if (svc_ctx == NULL) {
1442 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1444 * normal import, determine flavor from rule set, except
1445 * for mgc the flavor is predetermined.
1447 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1448 sf = cliobd->cl_flvr_mgc;
1450 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1452 &cliobd->cl_target_uuid,
1455 sp = imp->imp_obd->u.cli.cl_sp_me;
1457 /* reverse import, determine flavor from incoming reqeust */
1460 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1461 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1462 PTLRPC_SEC_FL_ROOTONLY;
1464 sp = sptlrpc_target_sec_part(imp->imp_obd);
1467 sec = sptlrpc_import_sec_ref(imp);
1471 if (flavor_equal(&sf, &sec->ps_flvr))
1474 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1475 imp->imp_obd->obd_name,
1476 obd_uuid2str(&conn->c_remote_uuid),
1477 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1478 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1479 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1480 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1481 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1482 imp->imp_obd->obd_name,
1483 obd_uuid2str(&conn->c_remote_uuid),
1484 LNET_NIDNET(conn->c_self),
1485 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1488 mutex_lock(&imp->imp_sec_mutex);
1490 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1492 sptlrpc_import_sec_install(imp, newsec);
1494 CERROR("import %s->%s: failed to create new sec\n",
1495 imp->imp_obd->obd_name,
1496 obd_uuid2str(&conn->c_remote_uuid));
1500 mutex_unlock(&imp->imp_sec_mutex);
1502 sptlrpc_sec_put(sec);
1506 void sptlrpc_import_sec_put(struct obd_import *imp)
1509 sptlrpc_sec_kill(imp->imp_sec);
1511 sptlrpc_sec_put(imp->imp_sec);
1512 imp->imp_sec = NULL;
1516 static void import_flush_ctx_common(struct obd_import *imp,
1517 uid_t uid, int grace, int force)
1519 struct ptlrpc_sec *sec;
1524 sec = sptlrpc_import_sec_ref(imp);
1528 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1529 sptlrpc_sec_put(sec);
1532 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1534 /* it's important to use grace mode, see explain in
1535 * sptlrpc_req_refresh_ctx() */
1536 import_flush_ctx_common(imp, 0, 1, 1);
1539 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1541 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1544 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1546 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1548 import_flush_ctx_common(imp, -1, 1, 1);
1550 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1553 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1554 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1556 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1558 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1559 struct ptlrpc_sec_policy *policy;
1563 LASSERT(ctx->cc_sec);
1564 LASSERT(ctx->cc_sec->ps_policy);
1565 LASSERT(req->rq_reqmsg == NULL);
1566 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1568 policy = ctx->cc_sec->ps_policy;
1569 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1571 LASSERT(req->rq_reqmsg);
1572 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1574 /* zeroing preallocated buffer */
1576 memset(req->rq_reqmsg, 0, msgsize);
1583 * Used by ptlrpc client to free request buffer of \a req. After this
1584 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1586 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1588 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1589 struct ptlrpc_sec_policy *policy;
1592 LASSERT(ctx->cc_sec);
1593 LASSERT(ctx->cc_sec->ps_policy);
1594 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1596 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1599 policy = ctx->cc_sec->ps_policy;
1600 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1601 req->rq_reqmsg = NULL;
1605 * NOTE caller must guarantee the buffer size is enough for the enlargement
1607 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1608 int segment, int newsize)
1611 int oldsize, oldmsg_size, movesize;
1613 LASSERT(segment < msg->lm_bufcount);
1614 LASSERT(msg->lm_buflens[segment] <= newsize);
1616 if (msg->lm_buflens[segment] == newsize)
1619 /* nothing to do if we are enlarging the last segment */
1620 if (segment == msg->lm_bufcount - 1) {
1621 msg->lm_buflens[segment] = newsize;
1625 oldsize = msg->lm_buflens[segment];
1627 src = lustre_msg_buf(msg, segment + 1, 0);
1628 msg->lm_buflens[segment] = newsize;
1629 dst = lustre_msg_buf(msg, segment + 1, 0);
1630 msg->lm_buflens[segment] = oldsize;
1632 /* move from segment + 1 to end segment */
1633 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1634 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1635 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1636 LASSERT(movesize >= 0);
1639 memmove(dst, src, movesize);
1641 /* note we don't clear the ares where old data live, not secret */
1643 /* finally set new segment size */
1644 msg->lm_buflens[segment] = newsize;
1646 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1649 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1650 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1651 * preserved after the enlargement. this must be called after original request
1652 * buffer being allocated.
1654 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1655 * so caller should refresh its local pointers if needed.
1657 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1658 const struct req_msg_field *field,
1661 struct req_capsule *pill = &req->rq_pill;
1662 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1663 struct ptlrpc_sec_cops *cops;
1664 struct lustre_msg *msg = req->rq_reqmsg;
1665 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1669 LASSERT(msg->lm_bufcount > segment);
1670 LASSERT(msg->lm_buflens[segment] <= newsize);
1672 if (msg->lm_buflens[segment] == newsize)
1675 cops = ctx->cc_sec->ps_policy->sp_cops;
1676 LASSERT(cops->enlarge_reqbuf);
1677 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1679 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1682 * Used by ptlrpc client to allocate reply buffer of \a req.
1684 * \note After this, req->rq_repmsg is still not accessible.
1686 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1688 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1689 struct ptlrpc_sec_policy *policy;
1693 LASSERT(ctx->cc_sec);
1694 LASSERT(ctx->cc_sec->ps_policy);
1699 policy = ctx->cc_sec->ps_policy;
1700 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1704 * Used by ptlrpc client to free reply buffer of \a req. After this
1705 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1707 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1709 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1710 struct ptlrpc_sec_policy *policy;
1714 LASSERT(ctx->cc_sec);
1715 LASSERT(ctx->cc_sec->ps_policy);
1716 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1718 if (req->rq_repbuf == NULL)
1720 LASSERT(req->rq_repbuf_len);
1722 policy = ctx->cc_sec->ps_policy;
1723 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1724 req->rq_repmsg = NULL;
1727 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1729 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1730 struct ptlrpc_cli_ctx *ctx)
1732 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1734 if (!policy->sp_cops->install_rctx)
1736 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1739 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1740 struct ptlrpc_svc_ctx *ctx)
1742 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1744 if (!policy->sp_sops->install_rctx)
1746 return policy->sp_sops->install_rctx(imp, ctx);
1749 /****************************************
1750 * server side security *
1751 ****************************************/
1753 static int flavor_allowed(struct sptlrpc_flavor *exp,
1754 struct ptlrpc_request *req)
1756 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1758 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1761 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1762 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1763 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1764 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1770 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1773 * Given an export \a exp, check whether the flavor of incoming \a req
1774 * is allowed by the export \a exp. Main logic is about taking care of
1775 * changing configurations. Return 0 means success.
1777 int sptlrpc_target_export_check(struct obd_export *exp,
1778 struct ptlrpc_request *req)
1780 struct sptlrpc_flavor flavor;
1785 /* client side export has no imp_reverse, skip
1786 * FIXME maybe we should check flavor this as well??? */
1787 if (exp->exp_imp_reverse == NULL)
1790 /* don't care about ctx fini rpc */
1791 if (req->rq_ctx_fini)
1794 spin_lock(&exp->exp_lock);
1796 /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1797 * the first req with the new flavor, then treat it as current flavor,
1798 * adapt reverse sec according to it.
1799 * note the first rpc with new flavor might not be with root ctx, in
1800 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1. */
1801 if (unlikely(exp->exp_flvr_changed) &&
1802 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1803 /* make the new flavor as "current", and old ones as
1804 * about-to-expire */
1805 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1806 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1807 flavor = exp->exp_flvr_old[1];
1808 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1809 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1810 exp->exp_flvr_old[0] = exp->exp_flvr;
1811 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1812 EXP_FLVR_UPDATE_EXPIRE;
1813 exp->exp_flvr = flavor;
1815 /* flavor change finished */
1816 exp->exp_flvr_changed = 0;
1817 LASSERT(exp->exp_flvr_adapt == 1);
1819 /* if it's gss, we only interested in root ctx init */
1820 if (req->rq_auth_gss &&
1821 !(req->rq_ctx_init &&
1822 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1823 req->rq_auth_usr_ost))) {
1824 spin_unlock(&exp->exp_lock);
1825 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1826 req->rq_auth_gss, req->rq_ctx_init,
1827 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1828 req->rq_auth_usr_ost);
1832 exp->exp_flvr_adapt = 0;
1833 spin_unlock(&exp->exp_lock);
1835 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1836 req->rq_svc_ctx, &flavor);
1839 /* if it equals to the current flavor, we accept it, but need to
1840 * dealing with reverse sec/ctx */
1841 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1842 /* most cases should return here, we only interested in
1843 * gss root ctx init */
1844 if (!req->rq_auth_gss || !req->rq_ctx_init ||
1845 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
1846 !req->rq_auth_usr_ost)) {
1847 spin_unlock(&exp->exp_lock);
1851 /* if flavor just changed, we should not proceed, just leave
1852 * it and current flavor will be discovered and replaced
1853 * shortly, and let _this_ rpc pass through */
1854 if (exp->exp_flvr_changed) {
1855 LASSERT(exp->exp_flvr_adapt);
1856 spin_unlock(&exp->exp_lock);
1860 if (exp->exp_flvr_adapt) {
1861 exp->exp_flvr_adapt = 0;
1862 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
1863 exp, exp->exp_flvr.sf_rpc,
1864 exp->exp_flvr_old[0].sf_rpc,
1865 exp->exp_flvr_old[1].sf_rpc);
1866 flavor = exp->exp_flvr;
1867 spin_unlock(&exp->exp_lock);
1869 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1873 CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
1874 "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
1875 exp->exp_flvr_old[0].sf_rpc,
1876 exp->exp_flvr_old[1].sf_rpc);
1877 spin_unlock(&exp->exp_lock);
1879 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
1884 if (exp->exp_flvr_expire[0]) {
1885 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
1886 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
1887 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
1888 exp->exp_flvr.sf_rpc,
1889 exp->exp_flvr_old[0].sf_rpc,
1890 exp->exp_flvr_old[1].sf_rpc,
1891 (s64)(exp->exp_flvr_expire[0] -
1892 ktime_get_real_seconds()));
1893 spin_unlock(&exp->exp_lock);
1897 CDEBUG(D_SEC, "mark middle expired\n");
1898 exp->exp_flvr_expire[0] = 0;
1900 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
1901 exp->exp_flvr.sf_rpc,
1902 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1903 req->rq_flvr.sf_rpc);
1906 /* now it doesn't match the current flavor, the only chance we can
1907 * accept it is match the old flavors which is not expired. */
1908 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
1909 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
1910 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
1911 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
1913 exp->exp_flvr.sf_rpc,
1914 exp->exp_flvr_old[0].sf_rpc,
1915 exp->exp_flvr_old[1].sf_rpc,
1916 (s64)(exp->exp_flvr_expire[1] -
1917 ktime_get_real_seconds()));
1918 spin_unlock(&exp->exp_lock);
1922 CDEBUG(D_SEC, "mark oldest expired\n");
1923 exp->exp_flvr_expire[1] = 0;
1925 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
1926 exp, exp->exp_flvr.sf_rpc,
1927 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
1928 req->rq_flvr.sf_rpc);
1930 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
1931 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
1932 exp->exp_flvr_old[1].sf_rpc);
1935 spin_unlock(&exp->exp_lock);
1937 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
1938 exp, exp->exp_obd->obd_name,
1939 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
1940 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
1941 req->rq_flvr.sf_rpc,
1942 exp->exp_flvr.sf_rpc,
1943 exp->exp_flvr_old[0].sf_rpc,
1944 exp->exp_flvr_expire[0] ?
1945 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
1946 exp->exp_flvr_old[1].sf_rpc,
1947 exp->exp_flvr_expire[1] ?
1948 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
1951 EXPORT_SYMBOL(sptlrpc_target_export_check);
1953 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
1954 struct sptlrpc_rule_set *rset)
1956 struct obd_export *exp;
1957 struct sptlrpc_flavor new_flvr;
1961 spin_lock(&obd->obd_dev_lock);
1963 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1964 if (exp->exp_connection == NULL)
1967 /* note if this export had just been updated flavor
1968 * (exp_flvr_changed == 1), this will override the
1970 spin_lock(&exp->exp_lock);
1971 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
1972 exp->exp_connection->c_peer.nid,
1974 if (exp->exp_flvr_changed ||
1975 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
1976 exp->exp_flvr_old[1] = new_flvr;
1977 exp->exp_flvr_expire[1] = 0;
1978 exp->exp_flvr_changed = 1;
1979 exp->exp_flvr_adapt = 1;
1981 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
1982 exp, sptlrpc_part2name(exp->exp_sp_peer),
1983 exp->exp_flvr.sf_rpc,
1984 exp->exp_flvr_old[1].sf_rpc);
1986 spin_unlock(&exp->exp_lock);
1989 spin_unlock(&obd->obd_dev_lock);
1991 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
1993 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
1995 /* peer's claim is unreliable unless gss is being used */
1996 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
1999 switch (req->rq_sp_from) {
2001 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2002 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2003 svc_rc = SECSVC_DROP;
2007 if (!req->rq_auth_usr_mdt) {
2008 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2009 svc_rc = SECSVC_DROP;
2013 if (!req->rq_auth_usr_ost) {
2014 DEBUG_REQ(D_ERROR, req, "faked source OST");
2015 svc_rc = SECSVC_DROP;
2020 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2021 !req->rq_auth_usr_ost) {
2022 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2023 svc_rc = SECSVC_DROP;
2028 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2029 svc_rc = SECSVC_DROP;
2036 * Used by ptlrpc server, to perform transformation upon request message of
2037 * incoming \a req. This must be the first thing to do with an incoming
2038 * request in ptlrpc layer.
2040 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2041 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2042 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2043 * reply message has been prepared.
2044 * \retval SECSVC_DROP failed, this request should be dropped.
2046 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2048 struct ptlrpc_sec_policy *policy;
2049 struct lustre_msg *msg = req->rq_reqbuf;
2054 LASSERT(req->rq_reqmsg == NULL);
2055 LASSERT(req->rq_repmsg == NULL);
2056 LASSERT(req->rq_svc_ctx == NULL);
2058 req->rq_req_swab_mask = 0;
2060 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2063 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2067 CERROR("error unpacking request from %s x%llu\n",
2068 libcfs_id2str(req->rq_peer), req->rq_xid);
2069 RETURN(SECSVC_DROP);
2072 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2073 req->rq_sp_from = LUSTRE_SP_ANY;
2074 req->rq_auth_uid = -1; /* set to INVALID_UID */
2075 req->rq_auth_mapped_uid = -1;
2077 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2079 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2080 RETURN(SECSVC_DROP);
2083 LASSERT(policy->sp_sops->accept);
2084 rc = policy->sp_sops->accept(req);
2085 sptlrpc_policy_put(policy);
2086 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2087 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2090 * if it's not null flavor (which means embedded packing msg),
2091 * reset the swab mask for the comming inner msg unpacking.
2093 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2094 req->rq_req_swab_mask = 0;
2096 /* sanity check for the request source */
2097 rc = sptlrpc_svc_check_from(req, rc);
2102 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2103 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2104 * a buffer of \a msglen size.
2106 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2108 struct ptlrpc_sec_policy *policy;
2109 struct ptlrpc_reply_state *rs;
2113 LASSERT(req->rq_svc_ctx);
2114 LASSERT(req->rq_svc_ctx->sc_policy);
2116 policy = req->rq_svc_ctx->sc_policy;
2117 LASSERT(policy->sp_sops->alloc_rs);
2119 rc = policy->sp_sops->alloc_rs(req, msglen);
2120 if (unlikely(rc == -ENOMEM)) {
2121 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2122 if (svcpt->scp_service->srv_max_reply_size <
2123 msglen + sizeof(struct ptlrpc_reply_state)) {
2124 /* Just return failure if the size is too big */
2125 CERROR("size of message is too big (%zd), %d allowed\n",
2126 msglen + sizeof(struct ptlrpc_reply_state),
2127 svcpt->scp_service->srv_max_reply_size);
2131 /* failed alloc, try emergency pool */
2132 rs = lustre_get_emerg_rs(svcpt);
2136 req->rq_reply_state = rs;
2137 rc = policy->sp_sops->alloc_rs(req, msglen);
2139 lustre_put_emerg_rs(rs);
2140 req->rq_reply_state = NULL;
2145 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2151 * Used by ptlrpc server, to perform transformation upon reply message.
2153 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2154 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2156 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2158 struct ptlrpc_sec_policy *policy;
2162 LASSERT(req->rq_svc_ctx);
2163 LASSERT(req->rq_svc_ctx->sc_policy);
2165 policy = req->rq_svc_ctx->sc_policy;
2166 LASSERT(policy->sp_sops->authorize);
2168 rc = policy->sp_sops->authorize(req);
2169 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2175 * Used by ptlrpc server, to free reply_state.
2177 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2179 struct ptlrpc_sec_policy *policy;
2180 unsigned int prealloc;
2183 LASSERT(rs->rs_svc_ctx);
2184 LASSERT(rs->rs_svc_ctx->sc_policy);
2186 policy = rs->rs_svc_ctx->sc_policy;
2187 LASSERT(policy->sp_sops->free_rs);
2189 prealloc = rs->rs_prealloc;
2190 policy->sp_sops->free_rs(rs);
2193 lustre_put_emerg_rs(rs);
2197 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2199 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2202 atomic_inc(&ctx->sc_refcount);
2205 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2207 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2212 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2213 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2214 if (ctx->sc_policy->sp_sops->free_ctx)
2215 ctx->sc_policy->sp_sops->free_ctx(ctx);
2217 req->rq_svc_ctx = NULL;
2220 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2222 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2227 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2228 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2229 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2231 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2233 /****************************************
2235 ****************************************/
2238 * Perform transformation upon bulk data pointed by \a desc. This is called
2239 * before transforming the request message.
2241 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2242 struct ptlrpc_bulk_desc *desc)
2244 struct ptlrpc_cli_ctx *ctx;
2246 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2248 if (!req->rq_pack_bulk)
2251 ctx = req->rq_cli_ctx;
2252 if (ctx->cc_ops->wrap_bulk)
2253 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2256 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2259 * This is called after unwrap the reply message.
2260 * return nob of actual plain text size received, or error code.
2262 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2263 struct ptlrpc_bulk_desc *desc,
2266 struct ptlrpc_cli_ctx *ctx;
2269 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2271 if (!req->rq_pack_bulk)
2272 return desc->bd_nob_transferred;
2274 ctx = req->rq_cli_ctx;
2275 if (ctx->cc_ops->unwrap_bulk) {
2276 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2280 return desc->bd_nob_transferred;
2282 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2285 * This is called after unwrap the reply message.
2286 * return 0 for success or error code.
2288 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2289 struct ptlrpc_bulk_desc *desc)
2291 struct ptlrpc_cli_ctx *ctx;
2294 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2296 if (!req->rq_pack_bulk)
2299 ctx = req->rq_cli_ctx;
2300 if (ctx->cc_ops->unwrap_bulk) {
2301 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2307 * if everything is going right, nob should equals to nob_transferred.
2308 * in case of privacy mode, nob_transferred needs to be adjusted.
2310 if (desc->bd_nob != desc->bd_nob_transferred) {
2311 CERROR("nob %d doesn't match transferred nob %d\n",
2312 desc->bd_nob, desc->bd_nob_transferred);
2318 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2320 #ifdef HAVE_SERVER_SUPPORT
2322 * Performe transformation upon outgoing bulk read.
2324 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2325 struct ptlrpc_bulk_desc *desc)
2327 struct ptlrpc_svc_ctx *ctx;
2329 LASSERT(req->rq_bulk_read);
2331 if (!req->rq_pack_bulk)
2334 ctx = req->rq_svc_ctx;
2335 if (ctx->sc_policy->sp_sops->wrap_bulk)
2336 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2340 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2343 * Performe transformation upon incoming bulk write.
2345 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2346 struct ptlrpc_bulk_desc *desc)
2348 struct ptlrpc_svc_ctx *ctx;
2351 LASSERT(req->rq_bulk_write);
2354 * if it's in privacy mode, transferred should >= expected; otherwise
2355 * transferred should == expected.
2357 if (desc->bd_nob_transferred < desc->bd_nob ||
2358 (desc->bd_nob_transferred > desc->bd_nob &&
2359 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2360 SPTLRPC_BULK_SVC_PRIV)) {
2361 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2362 desc->bd_nob_transferred, desc->bd_nob);
2366 if (!req->rq_pack_bulk)
2369 ctx = req->rq_svc_ctx;
2370 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2371 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2373 CERROR("error unwrap bulk: %d\n", rc);
2376 /* return 0 to allow reply be sent */
2379 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2382 * Prepare buffers for incoming bulk write.
2384 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2385 struct ptlrpc_bulk_desc *desc)
2387 struct ptlrpc_svc_ctx *ctx;
2389 LASSERT(req->rq_bulk_write);
2391 if (!req->rq_pack_bulk)
2394 ctx = req->rq_svc_ctx;
2395 if (ctx->sc_policy->sp_sops->prep_bulk)
2396 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2400 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2402 #endif /* HAVE_SERVER_SUPPORT */
2404 /****************************************
2405 * user descriptor helpers *
2406 ****************************************/
2408 int sptlrpc_current_user_desc_size(void)
2412 ngroups = current_ngroups;
2414 if (ngroups > LUSTRE_MAX_GROUPS)
2415 ngroups = LUSTRE_MAX_GROUPS;
2416 return sptlrpc_user_desc_size(ngroups);
2418 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2420 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2422 struct ptlrpc_user_desc *pud;
2424 pud = lustre_msg_buf(msg, offset, 0);
2426 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2427 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2428 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2429 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2430 pud->pud_cap = cfs_curproc_cap_pack();
2431 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2434 if (pud->pud_ngroups > current_ngroups)
2435 pud->pud_ngroups = current_ngroups;
2436 #ifdef HAVE_GROUP_INFO_GID
2437 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2438 pud->pud_ngroups * sizeof(__u32));
2439 #else /* !HAVE_GROUP_INFO_GID */
2440 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2441 pud->pud_ngroups * sizeof(__u32));
2442 #endif /* HAVE_GROUP_INFO_GID */
2443 task_unlock(current);
2447 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2449 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2451 struct ptlrpc_user_desc *pud;
2454 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2459 __swab32s(&pud->pud_uid);
2460 __swab32s(&pud->pud_gid);
2461 __swab32s(&pud->pud_fsuid);
2462 __swab32s(&pud->pud_fsgid);
2463 __swab32s(&pud->pud_cap);
2464 __swab32s(&pud->pud_ngroups);
2467 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2468 CERROR("%u groups is too large\n", pud->pud_ngroups);
2472 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2473 msg->lm_buflens[offset]) {
2474 CERROR("%u groups are claimed but bufsize only %u\n",
2475 pud->pud_ngroups, msg->lm_buflens[offset]);
2480 for (i = 0; i < pud->pud_ngroups; i++)
2481 __swab32s(&pud->pud_groups[i]);
2486 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2488 /****************************************
2490 ****************************************/
2492 const char * sec2target_str(struct ptlrpc_sec *sec)
2494 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2496 if (sec_is_reverse(sec))
2498 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2500 EXPORT_SYMBOL(sec2target_str);
2503 * return true if the bulk data is protected
2505 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2507 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2508 case SPTLRPC_BULK_SVC_INTG:
2509 case SPTLRPC_BULK_SVC_PRIV:
2515 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2517 /****************************************
2518 * crypto API helper/alloc blkciper *
2519 ****************************************/
2521 /****************************************
2522 * initialize/finalize *
2523 ****************************************/
2525 int sptlrpc_init(void)
2529 rwlock_init(&policy_lock);
2531 rc = sptlrpc_gc_init();
2535 rc = sptlrpc_conf_init();
2539 rc = sptlrpc_enc_pool_init();
2543 rc = sptlrpc_null_init();
2547 rc = sptlrpc_plain_init();
2551 rc = sptlrpc_lproc_init();
2558 sptlrpc_plain_fini();
2560 sptlrpc_null_fini();
2562 sptlrpc_enc_pool_fini();
2564 sptlrpc_conf_fini();
2571 void sptlrpc_fini(void)
2573 sptlrpc_lproc_fini();
2574 sptlrpc_plain_fini();
2575 sptlrpc_null_fini();
2576 sptlrpc_enc_pool_fini();
2577 sptlrpc_conf_fini();