-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Eric Mei <ericm@clusterfs.com>
*/
-#ifndef EXPORT_SYMTAB
-#define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_SEC
-#include <libcfs/libcfs.h>
-#ifndef __KERNEL__
-#include <liblustre.h>
-#include <libcfs/list.h>
-#else
+#include <linux/user_namespace.h>
+#ifdef HAVE_UIDGID_HEADER
+# include <linux/uidgid.h>
+#endif
#include <linux/crypto.h>
#include <linux/key.h>
-#endif
+#include <libcfs/libcfs.h>
#include <obd.h>
#include <obd_class.h>
#include <obd_support.h>
* policy registers *
***********************************************/
-static rwlock_t policy_lock = RW_LOCK_UNLOCKED;
+static rwlock_t policy_lock;
static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
NULL,
};
if (number >= SPTLRPC_POLICY_MAX)
return -EINVAL;
- write_lock(&policy_lock);
+ write_lock(&policy_lock);
if (unlikely(policies[number])) {
- write_unlock(&policy_lock);
+ write_unlock(&policy_lock);
return -EALREADY;
}
policies[number] = policy;
- write_unlock(&policy_lock);
+ write_unlock(&policy_lock);
CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
return 0;
LASSERT(number < SPTLRPC_POLICY_MAX);
- write_lock(&policy_lock);
+ write_lock(&policy_lock);
if (unlikely(policies[number] == NULL)) {
- write_unlock(&policy_lock);
+ write_unlock(&policy_lock);
CERROR("%s: already unregistered\n", policy->sp_name);
return -EINVAL;
}
LASSERT(policies[number] == policy);
policies[number] = NULL;
- write_unlock(&policy_lock);
+ write_unlock(&policy_lock);
CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
return 0;
EXPORT_SYMBOL(sptlrpc_unregister_policy);
static
-struct ptlrpc_sec_policy * sptlrpc_rpcflavor2policy(__u16 flavor)
-{
- static DECLARE_MUTEX(load_mutex);
- static atomic_t loaded = ATOMIC_INIT(0);
- struct ptlrpc_sec_policy *policy;
- __u16 number = RPC_FLVR_POLICY(flavor), flag = 0;
-
- if (number >= SPTLRPC_POLICY_MAX)
- return NULL;
-
- while (1) {
- read_lock(&policy_lock);
- policy = policies[number];
- if (policy && !try_module_get(policy->sp_owner))
- policy = NULL;
- if (policy == NULL)
- flag = atomic_read(&loaded);
- read_unlock(&policy_lock);
-
- if (policy != NULL || flag != 0 ||
- number != SPTLRPC_POLICY_GSS)
- break;
-
- /* try to load gss module, once */
- mutex_down(&load_mutex);
- if (atomic_read(&loaded) == 0) {
- if (request_module("ptlrpc_gss") == 0)
- CWARN("module ptlrpc_gss loaded on demand\n");
- else
- CERROR("Unable to load module ptlrpc_gss\n");
-
- atomic_set(&loaded, 1);
- }
- mutex_up(&load_mutex);
- }
-
- return policy;
-}
-
-__u16 sptlrpc_name2rpcflavor(const char *name)
-{
- if (!strcmp(name, "null"))
- return SPTLRPC_FLVR_NULL;
- if (!strcmp(name, "plain"))
- return SPTLRPC_FLVR_PLAIN;
+struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
+{
+ static DEFINE_MUTEX(load_mutex);
+ static atomic_t loaded = ATOMIC_INIT(0);
+ struct ptlrpc_sec_policy *policy;
+ __u16 number = SPTLRPC_FLVR_POLICY(flavor);
+ __u16 flag = 0;
+
+ if (number >= SPTLRPC_POLICY_MAX)
+ return NULL;
+
+ while (1) {
+ read_lock(&policy_lock);
+ policy = policies[number];
+ if (policy && !try_module_get(policy->sp_owner))
+ policy = NULL;
+ if (policy == NULL)
+ flag = atomic_read(&loaded);
+ read_unlock(&policy_lock);
+
+ if (policy != NULL || flag != 0 ||
+ number != SPTLRPC_POLICY_GSS)
+ break;
+
+ /* try to load gss module, once */
+ mutex_lock(&load_mutex);
+ if (atomic_read(&loaded) == 0) {
+ if (request_module("ptlrpc_gss") == 0)
+ CDEBUG(D_SEC,
+ "module ptlrpc_gss loaded on demand\n");
+ else
+ CERROR("Unable to load module ptlrpc_gss\n");
+
+ atomic_set(&loaded, 1);
+ }
+ mutex_unlock(&load_mutex);
+ }
+
+ return policy;
+}
+
+__u32 sptlrpc_name2flavor_base(const char *name)
+{
+ if (!strcmp(name, "null"))
+ return SPTLRPC_FLVR_NULL;
+ if (!strcmp(name, "plain"))
+ return SPTLRPC_FLVR_PLAIN;
+ if (!strcmp(name, "gssnull"))
+ return SPTLRPC_FLVR_GSSNULL;
if (!strcmp(name, "krb5n"))
return SPTLRPC_FLVR_KRB5N;
if (!strcmp(name, "krb5a"))
return SPTLRPC_FLVR_KRB5I;
if (!strcmp(name, "krb5p"))
return SPTLRPC_FLVR_KRB5P;
+ if (!strcmp(name, "skn"))
+ return SPTLRPC_FLVR_SKN;
+ if (!strcmp(name, "ska"))
+ return SPTLRPC_FLVR_SKA;
+ if (!strcmp(name, "ski"))
+ return SPTLRPC_FLVR_SKI;
+ if (!strcmp(name, "skpi"))
+ return SPTLRPC_FLVR_SKPI;
- return SPTLRPC_FLVR_INVALID;
+ return SPTLRPC_FLVR_INVALID;
}
-EXPORT_SYMBOL(sptlrpc_name2rpcflavor);
+EXPORT_SYMBOL(sptlrpc_name2flavor_base);
-const char *sptlrpc_rpcflavor2name(__u16 flavor)
+const char *sptlrpc_flavor2name_base(__u32 flvr)
{
- switch (flavor) {
- case SPTLRPC_FLVR_NULL:
+ __u32 base = SPTLRPC_FLVR_BASE(flvr);
+
+ if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
return "null";
- case SPTLRPC_FLVR_PLAIN:
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
return "plain";
- case SPTLRPC_FLVR_KRB5N:
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
+ return "gssnull";
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
return "krb5n";
- case SPTLRPC_FLVR_KRB5A:
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
return "krb5a";
- case SPTLRPC_FLVR_KRB5I:
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
return "krb5i";
- case SPTLRPC_FLVR_KRB5P:
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
return "krb5p";
- default:
- CERROR("invalid rpc flavor 0x%x(p%u,s%u,v%u)\n", flavor,
- RPC_FLVR_POLICY(flavor), RPC_FLVR_MECH(flavor),
- RPC_FLVR_SVC(flavor));
- }
- return "unknown";
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
+ return "skn";
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
+ return "ska";
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
+ return "ski";
+ else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
+ return "skpi";
+
+ CERROR("invalid wire flavor 0x%x\n", flvr);
+ return "invalid";
+}
+EXPORT_SYMBOL(sptlrpc_flavor2name_base);
+
+char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
+ char *buf, int bufsize)
+{
+ if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
+ snprintf(buf, bufsize, "hash:%s",
+ sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
+ else
+ snprintf(buf, bufsize, "%s",
+ sptlrpc_flavor2name_base(sf->sf_rpc));
+
+ buf[bufsize - 1] = '\0';
+ return buf;
}
-EXPORT_SYMBOL(sptlrpc_rpcflavor2name);
+EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
-int sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
+char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
{
- char *bulk;
+ snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
- if (sf->sf_bulk_ciph != BULK_CIPH_ALG_NULL)
- bulk = "bulkp";
- else if (sf->sf_bulk_hash != BULK_HASH_ALG_NULL)
- bulk = "bulki";
- else
- bulk = "bulkn";
+ /*
+ * currently we don't support customized bulk specification for
+ * flavors other than plain
+ */
+ if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
+ char bspec[16];
- snprintf(buf, bufsize, "%s-%s:%s/%s",
- sptlrpc_rpcflavor2name(sf->sf_rpc), bulk,
- sptlrpc_get_hash_name(sf->sf_bulk_hash),
- sptlrpc_get_ciph_name(sf->sf_bulk_ciph));
- return 0;
+ bspec[0] = '-';
+ sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
+ strncat(buf, bspec, bufsize);
+ }
+
+ buf[bufsize - 1] = '\0';
+ return buf;
}
EXPORT_SYMBOL(sptlrpc_flavor2name);
+char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
+{
+ buf[0] = '\0';
+
+ if (flags & PTLRPC_SEC_FL_REVERSE)
+ strlcat(buf, "reverse,", bufsize);
+ if (flags & PTLRPC_SEC_FL_ROOTONLY)
+ strlcat(buf, "rootonly,", bufsize);
+ if (flags & PTLRPC_SEC_FL_UDESC)
+ strlcat(buf, "udesc,", bufsize);
+ if (flags & PTLRPC_SEC_FL_BULK)
+ strlcat(buf, "bulk,", bufsize);
+ if (buf[0] == '\0')
+ strlcat(buf, "-,", bufsize);
+
+ return buf;
+}
+EXPORT_SYMBOL(sptlrpc_secflags2str);
+
/**************************************************
* client context APIs *
**************************************************/
create = 0;
remove_dead = 0;
}
- } else {
- vcred.vc_uid = cfs_current()->uid;
- vcred.vc_gid = cfs_current()->gid;
- }
+ } else {
+ vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
+ vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
+ }
- return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
- create, remove_dead);
+ return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
+ remove_dead);
}
struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
{
- LASSERT(atomic_read(&ctx->cc_refcount) > 0);
- atomic_inc(&ctx->cc_refcount);
- return ctx;
+ atomic_inc(&ctx->cc_refcount);
+ return ctx;
}
EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
{
- struct ptlrpc_sec *sec = ctx->cc_sec;
+ struct ptlrpc_sec *sec = ctx->cc_sec;
- LASSERT(sec);
- LASSERT(atomic_read(&ctx->cc_refcount));
+ LASSERT(sec);
+ LASSERT_ATOMIC_POS(&ctx->cc_refcount);
- if (!atomic_dec_and_test(&ctx->cc_refcount))
- return;
+ if (!atomic_dec_and_test(&ctx->cc_refcount))
+ return;
- sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
+ sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
}
EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
-/*
- * expire the context immediately.
- * the caller must hold at least 1 ref on the ctx.
+/**
+ * Expire the client context immediately.
+ *
+ * \pre Caller must hold at least 1 reference on the \a ctx.
*/
void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
{
}
EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
+/**
+ * To wake up the threads who are waiting for this client context. Called
+ * after some status change happened on \a ctx.
+ */
void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
{
- struct ptlrpc_request *req, *next;
+ struct ptlrpc_request *req, *next;
- spin_lock(&ctx->cc_lock);
- list_for_each_entry_safe(req, next, &ctx->cc_req_list, rq_ctx_chain) {
- list_del_init(&req->rq_ctx_chain);
- ptlrpc_wake_client_req(req);
- }
- spin_unlock(&ctx->cc_lock);
+ spin_lock(&ctx->cc_lock);
+ list_for_each_entry_safe(req, next, &ctx->cc_req_list,
+ rq_ctx_chain) {
+ list_del_init(&req->rq_ctx_chain);
+ ptlrpc_client_wake_req(req);
+ }
+ spin_unlock(&ctx->cc_lock);
}
EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
return ctx->cc_ops->display(ctx, buf, bufsize);
}
-static int sptlrpc_import_sec_check_expire(struct obd_import *imp)
+static int import_sec_check_expire(struct obd_import *imp)
{
- int adapt = 0;
+ int adapt = 0;
- spin_lock(&imp->imp_lock);
- if (imp->imp_sec_expire &&
- imp->imp_sec_expire < cfs_time_current_sec()) {
- adapt = 1;
- imp->imp_sec_expire = 0;
- }
- spin_unlock(&imp->imp_lock);
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_sec_expire &&
+ imp->imp_sec_expire < ktime_get_real_seconds()) {
+ adapt = 1;
+ imp->imp_sec_expire = 0;
+ }
+ spin_unlock(&imp->imp_lock);
if (!adapt)
return 0;
CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
- return sptlrpc_import_sec_adapt(imp, NULL, 0);
+ return sptlrpc_import_sec_adapt(imp, NULL, NULL);
}
+/**
+ * Get and validate the client side ptlrpc security facilities from
+ * \a imp. There is a race condition on client reconnect when the import is
+ * being destroyed while there are outstanding client bound requests. In
+ * this case do not output any error messages if import secuity is not
+ * found.
+ *
+ * \param[in] imp obd import associated with client
+ * \param[out] sec client side ptlrpc security
+ *
+ * \retval 0 if security retrieved successfully
+ * \retval -ve errno if there was a problem
+ */
+static int import_sec_validate_get(struct obd_import *imp,
+ struct ptlrpc_sec **sec)
+{
+ int rc;
+
+ if (unlikely(imp->imp_sec_expire)) {
+ rc = import_sec_check_expire(imp);
+ if (rc)
+ return rc;
+ }
+
+ *sec = sptlrpc_import_sec_ref(imp);
+ if (*sec == NULL) {
+ CERROR("import %p (%s) with no sec\n",
+ imp, ptlrpc_import_state_name(imp->imp_state));
+ return -EACCES;
+ }
+
+ if (unlikely((*sec)->ps_dying)) {
+ CERROR("attempt to use dying sec %p\n", sec);
+ sptlrpc_sec_put(*sec);
+ return -EACCES;
+ }
+
+ return 0;
+}
+
+/**
+ * Given a \a req, find or allocate an appropriate context for it.
+ * \pre req->rq_cli_ctx == NULL.
+ *
+ * \retval 0 succeed, and req->rq_cli_ctx is set.
+ * \retval -ev error number, and req->rq_cli_ctx == NULL.
+ */
int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
{
struct obd_import *imp = req->rq_import;
LASSERT(!req->rq_cli_ctx);
LASSERT(imp);
- if (unlikely(imp->imp_sec_expire)) {
- rc = sptlrpc_import_sec_check_expire(imp);
- if (rc)
- RETURN(rc);
- }
-
- sec = sptlrpc_import_sec_ref(imp);
- if (sec == NULL) {
- CERROR("import %p (%s) with no ptlrpc_sec\n",
- imp, ptlrpc_import_state_name(imp->imp_state));
- RETURN(-EACCES);
- }
-
- if (unlikely(sec->ps_dying)) {
- CERROR("attempt to use dying sec %p\n", sec);
- return -EACCES;
- }
+ rc = import_sec_validate_get(imp, &sec);
+ if (rc)
+ RETURN(rc);
req->rq_cli_ctx = get_my_ctx(sec);
sptlrpc_sec_put(sec);
- if (!req->rq_cli_ctx) {
- CERROR("req %p: fail to get context\n", req);
- RETURN(-ENOMEM);
- }
+ if (!req->rq_cli_ctx) {
+ CERROR("req %p: fail to get context\n", req);
+ RETURN(-ECONNREFUSED);
+ }
RETURN(0);
}
-/*
- * if @sync == 0, this function should return quickly without sleep;
- * otherwise might trigger ctx destroying rpc to server.
+/**
+ * Drop the context for \a req.
+ * \pre req->rq_cli_ctx != NULL.
+ * \post req->rq_cli_ctx == NULL.
+ *
+ * If \a sync == 0, this function should return quickly without sleep;
+ * otherwise it might trigger and wait for the whole process of sending
+ * an context-destroying rpc to server.
*/
void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
{
/* request might be asked to release earlier while still
* in the context waiting list.
*/
- if (!list_empty(&req->rq_ctx_chain)) {
- spin_lock(&req->rq_cli_ctx->cc_lock);
- list_del_init(&req->rq_ctx_chain);
- spin_unlock(&req->rq_cli_ctx->cc_lock);
+ if (!list_empty(&req->rq_ctx_chain)) {
+ spin_lock(&req->rq_cli_ctx->cc_lock);
+ list_del_init(&req->rq_ctx_chain);
+ spin_unlock(&req->rq_cli_ctx->cc_lock);
}
sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
struct ptlrpc_cli_ctx *newctx)
{
struct sptlrpc_flavor old_flvr;
- char *reqmsg;
+ char *reqmsg = NULL; /* to workaround old gcc */
int reqmsg_size;
- int rc;
-
- if (likely(oldctx->cc_sec == newctx->cc_sec))
- return 0;
+ int rc = 0;
LASSERT(req->rq_reqmsg);
LASSERT(req->rq_reqlen);
LASSERT(req->rq_replen);
- CWARN("req %p: switch ctx %p -> %p, switch sec %p(%s) -> %p(%s)\n",
- req, oldctx, newctx,
- oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
- newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
+ CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
+ "switch sec %p(%s) -> %p(%s)\n", req,
+ oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
+ newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
+ oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
+ newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
/* save flavor */
old_flvr = req->rq_flvr;
/* save request message */
reqmsg_size = req->rq_reqlen;
- OBD_ALLOC(reqmsg, reqmsg_size);
- if (reqmsg == NULL)
- return -ENOMEM;
- memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
+ if (reqmsg_size != 0) {
+ OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
+ if (reqmsg == NULL)
+ return -ENOMEM;
+ memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
+ }
/* release old req/rep buf */
req->rq_cli_ctx = oldctx;
/* alloc new request buffer
* we don't need to alloc reply buffer here, leave it to the
- * rest procedure of ptlrpc
- */
- rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
- if (!rc) {
- LASSERT(req->rq_reqmsg);
- memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
- } else {
- CWARN("failed to alloc reqbuf: %d\n", rc);
- req->rq_flvr = old_flvr;
- }
+ * rest procedure of ptlrpc */
+ if (reqmsg_size != 0) {
+ rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
+ if (!rc) {
+ LASSERT(req->rq_reqmsg);
+ memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
+ } else {
+ CWARN("failed to alloc reqbuf: %d\n", rc);
+ req->rq_flvr = old_flvr;
+ }
- OBD_FREE(reqmsg, reqmsg_size);
+ OBD_FREE_LARGE(reqmsg, reqmsg_size);
+ }
return rc;
}
-/*
- * request must have a context. in any case of failure, restore the
- * restore the old one. a request must have a ctx.
+/**
+ * If current context of \a req is dead somehow, e.g. we just switched flavor
+ * thus marked original contexts dead, we'll find a new context for it. if
+ * no switch is needed, \a req will end up with the same context.
+ *
+ * \note a request must have a context, to keep other parts of code happy.
+ * In any case of failure during the switching, we must restore the old one.
*/
int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
{
ENTRY;
LASSERT(oldctx);
- LASSERT(test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags));
sptlrpc_cli_ctx_get(oldctx);
sptlrpc_req_put_ctx(req, 0);
newctx = req->rq_cli_ctx;
LASSERT(newctx);
- if (unlikely(newctx == oldctx)) {
+ if (unlikely(newctx == oldctx &&
+ test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
/*
- * still get the old ctx, usually means system busy
+ * still get the old dead ctx, usually means system too busy
+ */
+ CDEBUG(D_SEC,
+ "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
+ newctx, newctx->cc_flags);
+
+ set_current_state(TASK_INTERRUPTIBLE);
+ schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
+ } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
+ == 0)) {
+ /*
+ * new ctx not up to date yet
+ */
+ CDEBUG(D_SEC,
+ "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
+ newctx, newctx->cc_flags);
+ } else {
+ /*
+ * it's possible newctx == oldctx if we're switching
+ * subflavor with the same sec.
*/
- CWARN("ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
- newctx, newctx->cc_flags);
-
- cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE, HZ);
- } else {
rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
if (rc) {
/* restore old ctx */
/* conn_cnt is needed in expire_one_request */
lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
- rc = ptlrpc_expire_one_request(req);
+ rc = ptlrpc_expire_one_request(req, 1);
/* if we started recovery, we should mark this ctx dead; otherwise
* in case of lgssd died nobody would retire this ctx, following
* connecting will still find the same ctx thus cause deadlock.
static
void ctx_refresh_interrupt(void *data)
{
- struct ptlrpc_request *req = data;
+ struct ptlrpc_request *req = data;
- spin_lock(&req->rq_lock);
- req->rq_intr = 1;
- spin_unlock(&req->rq_lock);
+ spin_lock(&req->rq_lock);
+ req->rq_intr = 1;
+ spin_unlock(&req->rq_lock);
}
static
void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
{
- spin_lock(&ctx->cc_lock);
- if (!list_empty(&req->rq_ctx_chain))
- list_del_init(&req->rq_ctx_chain);
- spin_unlock(&ctx->cc_lock);
+ spin_lock(&ctx->cc_lock);
+ if (!list_empty(&req->rq_ctx_chain))
+ list_del_init(&req->rq_ctx_chain);
+ spin_unlock(&ctx->cc_lock);
}
-/*
- * the status of context could be subject to be changed by other threads at any
- * time. we allow this race. but once we return with 0, the caller will
- * suppose it's uptodated and keep using it until the owning rpc is done.
+/**
+ * To refresh the context of \req, if it's not up-to-date.
+ * \param timeout
+ * - < 0: don't wait
+ * - = 0: wait until success or fatal error occur
+ * - > 0: timeout value (in seconds)
*
- * @timeout:
- * < 0 - don't wait
- * = 0 - wait until success or fatal error occur
- * > 0 - timeout value
+ * The status of the context could be subject to be changed by other threads
+ * at any time. We allow this race, but once we return with 0, the caller will
+ * suppose it's uptodated and keep using it until the owning rpc is done.
*
- * return 0 only if the context is uptodated.
+ * \retval 0 only if the context is uptodated.
+ * \retval -ev error number.
*/
int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
+ struct ptlrpc_sec *sec;
struct l_wait_info lwi;
int rc;
ENTRY;
LASSERT(ctx);
+ if (req->rq_ctx_init || req->rq_ctx_fini)
+ RETURN(0);
+
/*
* during the process a request's context might change type even
- * (e.g. from gss ctx to plain ctx), so each loop we need to re-check
+ * (e.g. from gss ctx to null ctx), so each loop we need to re-check
* everything
*/
again:
- /* skip special ctxs */
- if (cli_ctx_is_eternal(ctx) || req->rq_ctx_init || req->rq_ctx_fini)
+ rc = import_sec_validate_get(req->rq_import, &sec);
+ if (rc)
+ RETURN(rc);
+
+ if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
+ CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
+ req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
+ req_off_ctx_list(req, ctx);
+ sptlrpc_req_replace_dead_ctx(req);
+ ctx = req->rq_cli_ctx;
+ }
+ sptlrpc_sec_put(sec);
+
+ if (cli_ctx_is_eternal(ctx))
RETURN(0);
- if (test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags)) {
+ if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
LASSERT(ctx->cc_ops->refresh);
ctx->cc_ops->refresh(ctx);
}
- LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
+ LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
LASSERT(ctx->cc_ops->validate);
if (ctx->cc_ops->validate(ctx) == 0) {
RETURN(0);
}
- if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
- req->rq_err = 1;
+ if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
+ spin_lock(&req->rq_lock);
+ req->rq_err = 1;
+ spin_unlock(&req->rq_lock);
req_off_ctx_list(req, ctx);
RETURN(-EPERM);
}
- /* This is subtle. For resent message we have to keep original
- * context to survive following situation:
- * 1. the request sent to server
- * 2. recovery was kick start
- * 3. recovery finished, the request marked as resent
- * 4. resend the request
- * 5. old reply from server received (because xid is the same)
- * 6. verify reply (has to be success)
- * 7. new reply from server received, lnet drop it
+ /*
+ * There's a subtle issue for resending RPCs, suppose following
+ * situation:
+ * 1. the request was sent to server.
+ * 2. recovery was kicked start, after finished the request was
+ * marked as resent.
+ * 3. resend the request.
+ * 4. old reply from server received, we accept and verify the reply.
+ * this has to be success, otherwise the error will be aware
+ * by application.
+ * 5. new reply from server received, dropped by LNet.
*
- * Note we can't simply change xid for resent request because
- * server reply on it for reply reconstruction.
+ * Note the xid of old & new request is the same. We can't simply
+ * change xid for the resent request because the server replies on
+ * it for reply reconstruction.
*
* Commonly the original context should be uptodate because we
- * have a expiry nice time; And server will keep their half part
- * context because we at least hold a ref of old context which
- * prevent the context detroy RPC be sent. So server still can
- * accept the request and finish RPC. Two cases:
- * 1. If server side context has been trimed, a NO_CONTEXT will
+ * have an expiry nice time; server will keep its context because
+ * we at least hold a ref of old context which prevent context
+ * from destroying RPC being sent. So server still can accept the
+ * request and finish the RPC. But if that's not the case:
+ * 1. If server side context has been trimmed, a NO_CONTEXT will
* be returned, gss_cli_ctx_verify/unseal will switch to new
* context by force.
* 2. Current context never be refreshed, then we are fine: we
* never really send request with old context before.
*/
- if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
- unlikely(req->rq_reqmsg) &&
- lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
- req_off_ctx_list(req, ctx);
- RETURN(0);
- }
-
- if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
- rc = sptlrpc_req_replace_dead_ctx(req);
- if (rc) {
- LASSERT(ctx == req->rq_cli_ctx);
- CERROR("req %p: failed to replace dead ctx %p: %d\n",
- req, ctx, rc);
- req->rq_err = 1;
- LASSERT(list_empty(&req->rq_ctx_chain));
+ if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
+ unlikely(req->rq_reqmsg) &&
+ lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
+ req_off_ctx_list(req, ctx);
+ RETURN(0);
+ }
+
+ if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
+ req_off_ctx_list(req, ctx);
+ /*
+ * don't switch ctx if import was deactivated
+ */
+ if (req->rq_import->imp_deactive) {
+ spin_lock(&req->rq_lock);
+ req->rq_err = 1;
+ spin_unlock(&req->rq_lock);
+ RETURN(-EINTR);
+ }
+
+ rc = sptlrpc_req_replace_dead_ctx(req);
+ if (rc) {
+ LASSERT(ctx == req->rq_cli_ctx);
+ CERROR("req %p: failed to replace dead ctx %p: %d\n",
+ req, ctx, rc);
+ spin_lock(&req->rq_lock);
+ req->rq_err = 1;
+ spin_unlock(&req->rq_lock);
RETURN(rc);
}
- CWARN("req %p: replace dead ctx %p => ctx %p (%u->%s)\n",
- req, ctx, req->rq_cli_ctx,
- req->rq_cli_ctx->cc_vcred.vc_uid,
- sec2target_str(req->rq_cli_ctx->cc_sec));
-
ctx = req->rq_cli_ctx;
- LASSERT(list_empty(&req->rq_ctx_chain));
-
goto again;
}
- /* Now we're sure this context is during upcall, add myself into
+ /*
+ * Now we're sure this context is during upcall, add myself into
* waiting list
*/
- spin_lock(&ctx->cc_lock);
- if (list_empty(&req->rq_ctx_chain))
- list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
- spin_unlock(&ctx->cc_lock);
-
- if (timeout < 0) {
- RETURN(-EWOULDBLOCK);
- }
-
- /* Clear any flags that may be present from previous sends */
- LASSERT(req->rq_receiving_reply == 0);
- spin_lock(&req->rq_lock);
- req->rq_err = 0;
- req->rq_timedout = 0;
- req->rq_resend = 0;
- req->rq_restart = 0;
- spin_unlock(&req->rq_lock);
-
- lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout,
- ctx_refresh_interrupt, req);
- rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
+ spin_lock(&ctx->cc_lock);
+ if (list_empty(&req->rq_ctx_chain))
+ list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
+ spin_unlock(&ctx->cc_lock);
+
+ if (timeout < 0)
+ RETURN(-EWOULDBLOCK);
+
+ /* Clear any flags that may be present from previous sends */
+ LASSERT(req->rq_receiving_reply == 0);
+ spin_lock(&req->rq_lock);
+ req->rq_err = 0;
+ req->rq_timedout = 0;
+ req->rq_resend = 0;
+ req->rq_restart = 0;
+ spin_unlock(&req->rq_lock);
+
+ lwi = LWI_TIMEOUT_INTR(msecs_to_jiffies(timeout * MSEC_PER_SEC),
+ ctx_refresh_timeout,
+ ctx_refresh_interrupt, req);
+ rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
- /* following cases we could be here:
+ /*
+ * following cases could lead us here:
* - successfully refreshed;
- * - interruptted;
+ * - interrupted;
* - timedout, and we don't want recover from the failure;
* - timedout, and waked up upon recovery finished;
* - someone else mark this ctx dead by force;
- * - someone invalidate the req and call wake_client_req(),
+ * - someone invalidate the req and call ptlrpc_client_wake_req(),
* e.g. ptlrpc_abort_inflight();
*/
if (!cli_ctx_is_refreshed(ctx)) {
goto again;
}
-/*
- * Note this could be called in two situations:
+/**
+ * Initialize flavor settings for \a req, according to \a opcode.
+ *
+ * \note this could be called in two situations:
* - new request from ptlrpc_pre_req(), with proper @opcode
* - old request which changed ctx in the middle, with @opcode == 0
*/
LASSERT(req->rq_cli_ctx->cc_sec);
LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
- /* special security flags accoding to opcode */
+ /* special security flags according to opcode */
switch (opcode) {
case OST_READ:
+ case MDS_READPAGE:
+ case MGS_CONFIG_READ:
+ case OBD_IDX_READ:
req->rq_bulk_read = 1;
break;
case OST_WRITE:
+ case MDS_WRITEPAGE:
req->rq_bulk_write = 1;
break;
case SEC_CTX_INIT:
sec = req->rq_cli_ctx->cc_sec;
- spin_lock(&sec->ps_lock);
- req->rq_flvr = sec->ps_flvr;
- spin_unlock(&sec->ps_lock);
+ spin_lock(&sec->ps_lock);
+ req->rq_flvr = sec->ps_flvr;
+ spin_unlock(&sec->ps_lock);
/* force SVC_NULL for context initiation rpc, SVC_INTG for context
* destruction rpc */
if (unlikely(req->rq_ctx_init))
- rpc_flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
+ flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
else if (unlikely(req->rq_ctx_fini))
- rpc_flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
+ flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
/* user descriptor flag, null security can't do it anyway */
if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
/* bulk security flag */
if ((req->rq_bulk_read || req->rq_bulk_write) &&
- (req->rq_flvr.sf_bulk_ciph != BULK_CIPH_ALG_NULL ||
- req->rq_flvr.sf_bulk_hash != BULK_HASH_ALG_NULL))
+ sptlrpc_flavor_has_bulk(&req->rq_flvr))
req->rq_pack_bulk = 1;
}
void sptlrpc_request_out_callback(struct ptlrpc_request *req)
{
- if (RPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
+ if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
return;
LASSERT(req->rq_clrbuf);
req->rq_reqbuf_len = 0;
}
-/*
- * check whether current user have valid context for an import or not.
- * might repeatedly try in case of non-fatal errors.
- * return 0 on success, < 0 on failure
+/**
+ * Given an import \a imp, check whether current user has a valid context
+ * or not. We may create a new context and try to refresh it, and try
+ * repeatedly try in case of non-fatal errors. Return 0 means success.
*/
int sptlrpc_import_check_ctx(struct obd_import *imp)
{
- struct ptlrpc_sec *sec;
- struct ptlrpc_cli_ctx *ctx;
- struct ptlrpc_request *req = NULL;
- int rc;
- ENTRY;
+ struct ptlrpc_sec *sec;
+ struct ptlrpc_cli_ctx *ctx;
+ struct ptlrpc_request *req = NULL;
+ int rc;
+ ENTRY;
- might_sleep();
+ might_sleep();
- sec = sptlrpc_import_sec_ref(imp);
- ctx = get_my_ctx(sec);
- sptlrpc_sec_put(sec);
+ sec = sptlrpc_import_sec_ref(imp);
+ ctx = get_my_ctx(sec);
+ sptlrpc_sec_put(sec);
if (!ctx)
- RETURN(1);
+ RETURN(-ENOMEM);
if (cli_ctx_is_eternal(ctx) ||
ctx->cc_ops->validate(ctx) == 0) {
RETURN(0);
}
- OBD_ALLOC_PTR(req);
- if (!req)
- RETURN(-ENOMEM);
+ if (cli_ctx_is_error(ctx)) {
+ sptlrpc_cli_ctx_put(ctx, 1);
+ RETURN(-EACCES);
+ }
- spin_lock_init(&req->rq_lock);
- atomic_set(&req->rq_refcount, 10000);
- CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
- cfs_waitq_init(&req->rq_reply_waitq);
- req->rq_import = imp;
- req->rq_cli_ctx = ctx;
+ req = ptlrpc_request_cache_alloc(GFP_NOFS);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ ptlrpc_cli_req_init(req);
+ atomic_set(&req->rq_refcount, 10000);
+
+ req->rq_import = imp;
+ req->rq_flvr = sec->ps_flvr;
+ req->rq_cli_ctx = ctx;
rc = sptlrpc_req_refresh_ctx(req, 0);
- LASSERT(list_empty(&req->rq_ctx_chain));
+ LASSERT(list_empty(&req->rq_ctx_chain));
sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
- OBD_FREE_PTR(req);
+ ptlrpc_request_cache_free(req);
- RETURN(rc);
+ RETURN(rc);
}
+/**
+ * Used by ptlrpc client, to perform the pre-defined security transformation
+ * upon the request message of \a req. After this function called,
+ * req->rq_reqmsg is still accessible as clear text.
+ */
int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
RETURN(rc);
}
- switch (RPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
+ switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
case SPTLRPC_SVC_NULL:
case SPTLRPC_SVC_AUTH:
case SPTLRPC_SVC_INTG:
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
int rc;
- __u16 rpc_flvr;
ENTRY;
LASSERT(ctx);
LASSERT(req->rq_repdata);
LASSERT(req->rq_repmsg == NULL);
+ req->rq_rep_swab_mask = 0;
+
+ rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
+ switch (rc) {
+ case 1:
+ lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
+ case 0:
+ break;
+ default:
+ CERROR("failed unpack reply: x%llu\n", req->rq_xid);
+ RETURN(-EPROTO);
+ }
+
if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
CERROR("replied data length %d too small\n",
req->rq_repdata_len);
RETURN(-EPROTO);
}
- /* v2 message, check request/reply policy match */
- rpc_flvr = WIRE_FLVR_RPC(req->rq_repdata->lm_secflvr);
-
- if (req->rq_repdata->lm_magic == LUSTRE_MSG_MAGIC_V2_SWABBED)
- __swab16s(&rpc_flvr);
-
- if (RPC_FLVR_POLICY(rpc_flvr) !=
- RPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
- CERROR("request policy was %u while reply with %u\n",
- RPC_FLVR_POLICY(req->rq_flvr.sf_rpc),
- RPC_FLVR_POLICY(rpc_flvr));
+ if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
+ SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
+ CERROR("reply policy %u doesn't match request policy %u\n",
+ SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
+ SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
RETURN(-EPROTO);
}
- /* do nothing if it's null policy; otherwise unpack the
- * wrapper message */
- if (RPC_FLVR_POLICY(rpc_flvr) != SPTLRPC_POLICY_NULL &&
- lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len))
- RETURN(-EPROTO);
-
- switch (RPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
+ switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
case SPTLRPC_SVC_NULL:
case SPTLRPC_SVC_AUTH:
case SPTLRPC_SVC_INTG:
default:
LBUG();
}
-
LASSERT(rc || req->rq_repmsg || req->rq_resend);
+
+ if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
+ !req->rq_ctx_init)
+ req->rq_rep_swab_mask = 0;
RETURN(rc);
}
-/*
- * upon this be called, the reply buffer should have been un-posted,
- * so nothing is going to change.
+/**
+ * Used by ptlrpc client, to perform security transformation upon the reply
+ * message of \a req. After return successfully, req->rq_repmsg points to
+ * the reply message in clear text.
+ *
+ * \pre the reply buffer should have been un-posted from LNet, so nothing is
+ * going to change.
*/
int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
{
LASSERT(req->rq_repmsg == NULL);
LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
- if (req->rq_reply_off == 0) {
+ if (req->rq_reply_off == 0 &&
+ (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
CERROR("real reply with offset 0\n");
return -EPROTO;
}
}
/**
- * Upon called, the receive buffer might be still posted, so the reply data
- * might be changed at any time, no matter we're holding rq_lock or not. we
- * expect the rq_reply_off be 0, rq_nob_received is the early reply size.
+ * Used by ptlrpc client, to perform security transformation upon the early
+ * reply message of \a req. We expect the rq_reply_off is 0, and
+ * rq_nob_received is the early reply size.
+ *
+ * Because the receive buffer might be still posted, the reply data might be
+ * changed at any time, no matter we're holding rq_lock or not. For this reason
+ * we allocate a separate ptlrpc_request and reply buffer for early reply
+ * processing.
*
- * we allocate separate ptlrpc_request and reply buffer for early reply
- * processing, return 0 and @req_ret is a duplicated ptlrpc_request. caller
- * must call sptlrpc_cli_finish_early_reply() on the returned request to
- * release it. if anything goes wrong @req_ret will not be set.
+ * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
+ * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
+ * \a *req_ret to release it.
+ * \retval -ev error number, and \a req_ret will not be set.
*/
int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
- struct ptlrpc_request **req_ret)
-{
- struct ptlrpc_request *early_req;
- char *early_buf;
- int early_bufsz, early_size;
- int rc;
- ENTRY;
-
- OBD_ALLOC_PTR(early_req);
- if (early_req == NULL)
- RETURN(-ENOMEM);
-
- early_size = req->rq_nob_received;
- early_bufsz = size_roundup_power2(early_size);
- OBD_ALLOC(early_buf, early_bufsz);
- if (early_buf == NULL)
- GOTO(err_req, rc = -ENOMEM);
-
- /* sanity checkings and copy data out, do it inside spinlock */
- spin_lock(&req->rq_lock);
-
- if (req->rq_replied) {
- spin_unlock(&req->rq_lock);
- GOTO(err_buf, rc = -EALREADY);
- }
-
- LASSERT(req->rq_repbuf);
- LASSERT(req->rq_repdata == NULL);
- LASSERT(req->rq_repmsg == NULL);
-
- if (req->rq_reply_off != 0) {
- CERROR("early reply with offset %u\n", req->rq_reply_off);
- spin_unlock(&req->rq_lock);
- GOTO(err_buf, rc = -EPROTO);
- }
-
- if (req->rq_nob_received != early_size) {
- /* even another early arrived the size should be the same */
- CERROR("data size has changed from %u to %u\n",
- early_size, req->rq_nob_received);
- spin_unlock(&req->rq_lock);
- GOTO(err_buf, rc = -EINVAL);
- }
-
- if (req->rq_nob_received < sizeof(struct lustre_msg)) {
- CERROR("early reply length %d too small\n",
- req->rq_nob_received);
- spin_unlock(&req->rq_lock);
- GOTO(err_buf, rc = -EALREADY);
- }
-
- memcpy(early_buf, req->rq_repbuf, early_size);
- spin_unlock(&req->rq_lock);
+ struct ptlrpc_request **req_ret)
+{
+ struct ptlrpc_request *early_req;
+ char *early_buf;
+ int early_bufsz, early_size;
+ int rc;
+ ENTRY;
+
+ early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
+ if (early_req == NULL)
+ RETURN(-ENOMEM);
+
+ ptlrpc_cli_req_init(early_req);
+
+ early_size = req->rq_nob_received;
+ early_bufsz = size_roundup_power2(early_size);
+ OBD_ALLOC_LARGE(early_buf, early_bufsz);
+ if (early_buf == NULL)
+ GOTO(err_req, rc = -ENOMEM);
+
+ /* sanity checkings and copy data out, do it inside spinlock */
+ spin_lock(&req->rq_lock);
+
+ if (req->rq_replied) {
+ spin_unlock(&req->rq_lock);
+ GOTO(err_buf, rc = -EALREADY);
+ }
+
+ LASSERT(req->rq_repbuf);
+ LASSERT(req->rq_repdata == NULL);
+ LASSERT(req->rq_repmsg == NULL);
+
+ if (req->rq_reply_off != 0) {
+ CERROR("early reply with offset %u\n", req->rq_reply_off);
+ spin_unlock(&req->rq_lock);
+ GOTO(err_buf, rc = -EPROTO);
+ }
+
+ if (req->rq_nob_received != early_size) {
+ /* even another early arrived the size should be the same */
+ CERROR("data size has changed from %u to %u\n",
+ early_size, req->rq_nob_received);
+ spin_unlock(&req->rq_lock);
+ GOTO(err_buf, rc = -EINVAL);
+ }
+
+ if (req->rq_nob_received < sizeof(struct lustre_msg)) {
+ CERROR("early reply length %d too small\n",
+ req->rq_nob_received);
+ spin_unlock(&req->rq_lock);
+ GOTO(err_buf, rc = -EALREADY);
+ }
+
+ memcpy(early_buf, req->rq_repbuf, early_size);
+ spin_unlock(&req->rq_lock);
early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
early_req->rq_flvr = req->rq_flvr;
early_req->rq_repdata = (struct lustre_msg *) early_buf;
early_req->rq_repdata_len = early_size;
early_req->rq_early = 1;
+ early_req->rq_reqmsg = req->rq_reqmsg;
rc = do_cli_unwrap_reply(early_req);
if (rc) {
err_ctx:
sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
err_buf:
- OBD_FREE(early_buf, early_bufsz);
+ OBD_FREE_LARGE(early_buf, early_bufsz);
err_req:
- OBD_FREE_PTR(early_req);
- RETURN(rc);
+ ptlrpc_request_cache_free(early_req);
+ RETURN(rc);
}
+/**
+ * Used by ptlrpc client, to release a processed early reply \a early_req.
+ *
+ * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
+ */
void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
{
- LASSERT(early_req->rq_repbuf);
- LASSERT(early_req->rq_repdata);
- LASSERT(early_req->rq_repmsg);
+ LASSERT(early_req->rq_repbuf);
+ LASSERT(early_req->rq_repdata);
+ LASSERT(early_req->rq_repmsg);
- sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
- OBD_FREE(early_req->rq_repbuf, early_req->rq_repbuf_len);
- OBD_FREE_PTR(early_req);
+ sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
+ OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
+ ptlrpc_request_cache_free(early_req);
}
/**************************************************
int sptlrpc_get_next_secid(void)
{
- return atomic_inc_return(&sptlrpc_sec_id);
+ return atomic_inc_return(&sptlrpc_sec_id);
}
EXPORT_SYMBOL(sptlrpc_get_next_secid);
{
struct ptlrpc_sec_policy *policy = sec->ps_policy;
- LASSERT(atomic_read(&sec->ps_refcount) == 0);
- LASSERT(atomic_read(&sec->ps_nctx) == 0);
+ LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
+ LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
LASSERT(policy->sp_cops->destroy_sec);
CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
{
- LASSERT(atomic_read(&sec->ps_refcount) > 0);
+ LASSERT_ATOMIC_POS(&sec->ps_refcount);
if (sec->ps_policy->sp_cops->kill_sec) {
sec->ps_policy->sp_cops->kill_sec(sec);
struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
{
- if (sec) {
- LASSERT(atomic_read(&sec->ps_refcount) > 0);
- atomic_inc(&sec->ps_refcount);
- }
+ if (sec)
+ atomic_inc(&sec->ps_refcount);
- return sec;
+ return sec;
}
EXPORT_SYMBOL(sptlrpc_sec_get);
void sptlrpc_sec_put(struct ptlrpc_sec *sec)
{
- if (sec) {
- LASSERT(atomic_read(&sec->ps_refcount) > 0);
-
- if (atomic_dec_and_test(&sec->ps_refcount)) {
- LASSERT(atomic_read(&sec->ps_nctx) == 0);
+ if (sec) {
+ LASSERT_ATOMIC_POS(&sec->ps_refcount);
- sptlrpc_gc_del_sec(sec);
- sec_cop_destroy_sec(sec);
- }
- }
+ if (atomic_dec_and_test(&sec->ps_refcount)) {
+ sptlrpc_gc_del_sec(sec);
+ sec_cop_destroy_sec(sec);
+ }
+ }
}
EXPORT_SYMBOL(sptlrpc_sec_put);
/*
- * it's policy module responsible for taking refrence of import
+ * policy module is responsible for taking refrence of import
*/
static
struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
{
struct ptlrpc_sec_policy *policy;
struct ptlrpc_sec *sec;
+ char str[32];
ENTRY;
if (svc_ctx) {
CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
imp->imp_obd->obd_type->typ_name,
imp->imp_obd->obd_name,
- sptlrpc_rpcflavor2name(sf->sf_rpc));
+ sptlrpc_flavor2name(sf, str, sizeof(str)));
policy = sptlrpc_policy_get(svc_ctx->sc_policy);
sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
imp->imp_obd->obd_type->typ_name,
imp->imp_obd->obd_name,
- sptlrpc_rpcflavor2name(sf->sf_rpc));
+ sptlrpc_flavor2name(sf, str, sizeof(str)));
- policy = sptlrpc_rpcflavor2policy(sf->sf_rpc);
+ policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
if (!policy) {
CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
RETURN(NULL);
}
}
- sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
- if (sec) {
- atomic_inc(&sec->ps_refcount);
+ sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
+ if (sec) {
+ atomic_inc(&sec->ps_refcount);
- sec->ps_part = sp;
+ sec->ps_part = sp;
- if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
- sptlrpc_gc_add_sec(sec);
- } else {
- sptlrpc_policy_put(policy);
- }
+ if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
+ sptlrpc_gc_add_sec(sec);
+ } else {
+ sptlrpc_policy_put(policy);
+ }
- RETURN(sec);
+ RETURN(sec);
}
struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
{
- struct ptlrpc_sec *sec;
+ struct ptlrpc_sec *sec;
- spin_lock(&imp->imp_lock);
- sec = sptlrpc_sec_get(imp->imp_sec);
- spin_unlock(&imp->imp_lock);
+ spin_lock(&imp->imp_lock);
+ sec = sptlrpc_sec_get(imp->imp_sec);
+ spin_unlock(&imp->imp_lock);
- return sec;
+ return sec;
}
EXPORT_SYMBOL(sptlrpc_import_sec_ref);
static void sptlrpc_import_sec_install(struct obd_import *imp,
struct ptlrpc_sec *sec)
{
- struct ptlrpc_sec *old_sec;
+ struct ptlrpc_sec *old_sec;
- LASSERT(atomic_read(&sec->ps_refcount) > 0);
+ LASSERT_ATOMIC_POS(&sec->ps_refcount);
- spin_lock(&imp->imp_lock);
- old_sec = imp->imp_sec;
- imp->imp_sec = sec;
- spin_unlock(&imp->imp_lock);
+ spin_lock(&imp->imp_lock);
+ old_sec = imp->imp_sec;
+ imp->imp_sec = sec;
+ spin_unlock(&imp->imp_lock);
if (old_sec) {
sptlrpc_sec_kill(old_sec);
}
}
-static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
- struct ptlrpc_sec *sec,
- struct sptlrpc_flavor *sf)
-{
- if (sf->sf_bulk_ciph != sec->ps_flvr.sf_bulk_ciph ||
- sf->sf_bulk_hash != sec->ps_flvr.sf_bulk_hash) {
- CWARN("imp %p (%s->%s): changing bulk flavor %s/%s -> %s/%s\n",
- imp, imp->imp_obd->obd_name,
- obd_uuid2str(&imp->imp_connection->c_remote_uuid),
- sptlrpc_get_ciph_name(sec->ps_flvr.sf_bulk_ciph),
- sptlrpc_get_hash_name(sec->ps_flvr.sf_bulk_hash),
- sptlrpc_get_ciph_name(sf->sf_bulk_ciph),
- sptlrpc_get_hash_name(sf->sf_bulk_hash));
-
- spin_lock(&sec->ps_lock);
- sec->ps_flvr.sf_bulk_ciph = sf->sf_bulk_ciph;
- sec->ps_flvr.sf_bulk_hash = sf->sf_bulk_hash;
- spin_unlock(&sec->ps_lock);
- }
+static inline
+int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
+{
+ return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
+}
- if (!equi(sf->sf_flags & PTLRPC_SEC_FL_UDESC,
- sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC)) {
- CWARN("imp %p (%s->%s): %s shipping user descriptor\n",
- imp, imp->imp_obd->obd_name,
- obd_uuid2str(&imp->imp_connection->c_remote_uuid),
- (sf->sf_flags & PTLRPC_SEC_FL_UDESC) ? "start" : "stop");
-
- spin_lock(&sec->ps_lock);
- sec->ps_flvr.sf_flags &= ~PTLRPC_SEC_FL_UDESC;
- sec->ps_flvr.sf_flags |= sf->sf_flags & PTLRPC_SEC_FL_UDESC;
- spin_unlock(&sec->ps_lock);
- }
+static inline
+void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
+{
+ *dst = *src;
}
-/*
- * for normal import, @svc_ctx should be NULL and @rpc_flavor is ignored;
- * for reverse import, @svc_ctx and @rpc_flavor is from incoming request.
+/**
+ * To get an appropriate ptlrpc_sec for the \a imp, according to the current
+ * configuration. Upon called, imp->imp_sec may or may not be NULL.
+ *
+ * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
+ * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
*/
int sptlrpc_import_sec_adapt(struct obd_import *imp,
struct ptlrpc_svc_ctx *svc_ctx,
- __u16 rpc_flavor)
+ struct sptlrpc_flavor *flvr)
{
- struct ptlrpc_connection *conn;
- struct sptlrpc_flavor sf;
- struct ptlrpc_sec *sec, *newsec;
- enum lustre_sec_part sp;
- int rc;
+ struct ptlrpc_connection *conn;
+ struct sptlrpc_flavor sf;
+ struct ptlrpc_sec *sec, *newsec;
+ enum lustre_sec_part sp;
+ char str[24];
+ int rc = 0;
+ ENTRY;
- if (imp == NULL)
- return 0;
+ might_sleep();
- conn = imp->imp_connection;
+ if (imp == NULL)
+ RETURN(0);
- if (svc_ctx == NULL) {
- /* normal import, determine flavor from rule set */
- sptlrpc_rule_set_choose(&imp->imp_obd->u.cli.cl_sptlrpc_rset,
- LUSTRE_SP_ANY, conn->c_self, &sf);
+ conn = imp->imp_connection;
- sp = imp->imp_obd->u.cli.cl_sec_part;
+ if (svc_ctx == NULL) {
+ struct client_obd *cliobd = &imp->imp_obd->u.cli;
+ /*
+ * normal import, determine flavor from rule set, except
+ * for mgc the flavor is predetermined.
+ */
+ if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
+ sf = cliobd->cl_flvr_mgc;
+ else
+ sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
+ cliobd->cl_sp_to,
+ &cliobd->cl_target_uuid,
+ conn->c_self, &sf);
+
+ sp = imp->imp_obd->u.cli.cl_sp_me;
} else {
/* reverse import, determine flavor from incoming reqeust */
- sf.sf_rpc = rpc_flavor;
- sf.sf_bulk_ciph = BULK_CIPH_ALG_NULL;
- sf.sf_bulk_hash = BULK_HASH_ALG_NULL;
- sf.sf_flags = PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
+ sf = *flvr;
+
+ if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
+ sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
+ PTLRPC_SEC_FL_ROOTONLY;
sp = sptlrpc_target_sec_part(imp->imp_obd);
}
sec = sptlrpc_import_sec_ref(imp);
if (sec) {
- if (svc_ctx == NULL) {
- /* normal import, only check rpc flavor, if just bulk
- * flavor or flags changed, we can handle it on the fly
- * without switching sec. */
- if (sf.sf_rpc == sec->ps_flvr.sf_rpc) {
- sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
-
- rc = 0;
- goto out;
- }
- } else {
- /* reverse import, do not compare bulk flavor */
- if (sf.sf_rpc == sec->ps_flvr.sf_rpc) {
- rc = 0;
- goto out;
- }
- }
+ char str2[24];
- CWARN("%simport %p (%s%s%s): changing flavor "
- "(%s, %s/%s) -> (%s, %s/%s)\n",
- svc_ctx ? "reverse " : "",
- imp, imp->imp_obd->obd_name,
- svc_ctx == NULL ? "->" : "<-",
- obd_uuid2str(&conn->c_remote_uuid),
- sptlrpc_rpcflavor2name(sec->ps_flvr.sf_rpc),
- sptlrpc_get_hash_name(sec->ps_flvr.sf_bulk_hash),
- sptlrpc_get_ciph_name(sec->ps_flvr.sf_bulk_ciph),
- sptlrpc_rpcflavor2name(sf.sf_rpc),
- sptlrpc_get_hash_name(sf.sf_bulk_hash),
- sptlrpc_get_ciph_name(sf.sf_bulk_ciph));
- } else {
- CWARN("%simport %p (%s%s%s) netid %x: "
- "select initial flavor (%s, %s/%s)\n",
- svc_ctx == NULL ? "" : "reverse ",
- imp, imp->imp_obd->obd_name,
- svc_ctx == NULL ? "->" : "<-",
- obd_uuid2str(&conn->c_remote_uuid),
- LNET_NIDNET(conn->c_self),
- sptlrpc_rpcflavor2name(sf.sf_rpc),
- sptlrpc_get_hash_name(sf.sf_bulk_hash),
- sptlrpc_get_ciph_name(sf.sf_bulk_ciph));
+ if (flavor_equal(&sf, &sec->ps_flvr))
+ GOTO(out, rc);
+
+ CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
+ imp->imp_obd->obd_name,
+ obd_uuid2str(&conn->c_remote_uuid),
+ sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
+ sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
+ } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
+ SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
+ CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
+ imp->imp_obd->obd_name,
+ obd_uuid2str(&conn->c_remote_uuid),
+ LNET_NIDNET(conn->c_self),
+ sptlrpc_flavor2name(&sf, str, sizeof(str)));
}
- mutex_down(&imp->imp_sec_mutex);
+ mutex_lock(&imp->imp_sec_mutex);
newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
if (newsec) {
sptlrpc_import_sec_install(imp, newsec);
- rc = 0;
} else {
- CERROR("%simport %p (%s): failed to create new sec\n",
- svc_ctx == NULL ? "" : "reverse ",
- imp, obd_uuid2str(&conn->c_remote_uuid));
+ CERROR("import %s->%s: failed to create new sec\n",
+ imp->imp_obd->obd_name,
+ obd_uuid2str(&conn->c_remote_uuid));
rc = -EPERM;
}
- mutex_up(&imp->imp_sec_mutex);
-
+ mutex_unlock(&imp->imp_sec_mutex);
out:
sptlrpc_sec_put(sec);
- return 0;
+ RETURN(rc);
}
void sptlrpc_import_sec_put(struct obd_import *imp)
sptlrpc_sec_put(sec);
}
-void sptlrpc_import_inval_all_ctx(struct obd_import *imp)
-{
- /* use grace == 0 */
- import_flush_ctx_common(imp, -1, 0, 1);
-}
-
void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
{
/* it's important to use grace mode, see explain in
void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
{
- import_flush_ctx_common(imp, cfs_current()->uid, 1, 1);
+ import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
+ 1, 1);
}
EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
}
EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
-/*
- * when complete successfully, req->rq_reqmsg should point to the
- * right place.
+/**
+ * Used by ptlrpc client to allocate request buffer of \a req. Upon return
+ * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
*/
int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
{
int rc;
LASSERT(ctx);
- LASSERT(atomic_read(&ctx->cc_refcount));
LASSERT(ctx->cc_sec);
LASSERT(ctx->cc_sec->ps_policy);
LASSERT(req->rq_reqmsg == NULL);
+ LASSERT_ATOMIC_POS(&ctx->cc_refcount);
policy = ctx->cc_sec->ps_policy;
rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
return rc;
}
+/**
+ * Used by ptlrpc client to free request buffer of \a req. After this
+ * req->rq_reqmsg is set to NULL and should not be accessed anymore.
+ */
void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
struct ptlrpc_sec_policy *policy;
LASSERT(ctx);
- LASSERT(atomic_read(&ctx->cc_refcount));
LASSERT(ctx->cc_sec);
LASSERT(ctx->cc_sec->ps_policy);
+ LASSERT_ATOMIC_POS(&ctx->cc_refcount);
if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
return;
policy = ctx->cc_sec->ps_policy;
policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
+ req->rq_reqmsg = NULL;
}
/*
}
EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
-/*
- * enlarge @segment of upper message req->rq_reqmsg to @newsize, all data
- * will be preserved after enlargement. this must be called after rq_reqmsg has
- * been intialized at least.
+/**
+ * Used by ptlrpc client to enlarge the \a segment of request message pointed
+ * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
+ * preserved after the enlargement. this must be called after original request
+ * buffer being allocated.
*
- * caller's attention: upon return, rq_reqmsg and rq_reqlen might have
- * been changed.
+ * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
+ * so caller should refresh its local pointers if needed.
*/
int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
- int segment, int newsize)
+ const struct req_msg_field *field,
+ int newsize)
{
- struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
- struct ptlrpc_sec_cops *cops;
- struct lustre_msg *msg = req->rq_reqmsg;
+ struct req_capsule *pill = &req->rq_pill;
+ struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
+ struct ptlrpc_sec_cops *cops;
+ struct lustre_msg *msg = req->rq_reqmsg;
+ int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
LASSERT(ctx);
LASSERT(msg);
}
EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
+/**
+ * Used by ptlrpc client to allocate reply buffer of \a req.
+ *
+ * \note After this, req->rq_repmsg is still not accessible.
+ */
int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
ENTRY;
LASSERT(ctx);
- LASSERT(atomic_read(&ctx->cc_refcount));
LASSERT(ctx->cc_sec);
LASSERT(ctx->cc_sec->ps_policy);
RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
}
+/**
+ * Used by ptlrpc client to free reply buffer of \a req. After this
+ * req->rq_repmsg is set to NULL and should not be accessed anymore.
+ */
void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
{
struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
ENTRY;
LASSERT(ctx);
- LASSERT(atomic_read(&ctx->cc_refcount));
LASSERT(ctx->cc_sec);
LASSERT(ctx->cc_sec->ps_policy);
+ LASSERT_ATOMIC_POS(&ctx->cc_refcount);
if (req->rq_repbuf == NULL)
return;
policy = ctx->cc_sec->ps_policy;
policy->sp_cops->free_repbuf(ctx->cc_sec, req);
+ req->rq_repmsg = NULL;
EXIT;
}
+EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
struct ptlrpc_cli_ctx *ctx)
{
struct sptlrpc_flavor *flvr = &req->rq_flvr;
- if (exp->sf_rpc == flvr->sf_rpc)
+ if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
return 1;
if ((req->rq_ctx_init || req->rq_ctx_fini) &&
- RPC_FLVR_POLICY(exp->sf_rpc) == RPC_FLVR_POLICY(flvr->sf_rpc) &&
- RPC_FLVR_MECH(exp->sf_rpc) == RPC_FLVR_MECH(flvr->sf_rpc))
+ SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
+ SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
+ SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
return 1;
return 0;
#define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
+/**
+ * Given an export \a exp, check whether the flavor of incoming \a req
+ * is allowed by the export \a exp. Main logic is about taking care of
+ * changing configurations. Return 0 means success.
+ */
int sptlrpc_target_export_check(struct obd_export *exp,
struct ptlrpc_request *req)
{
if (req->rq_ctx_fini)
return 0;
- spin_lock(&exp->exp_lock);
+ spin_lock(&exp->exp_lock);
/* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
* the first req with the new flavor, then treat it as current flavor,
exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
exp->exp_flvr_old[0] = exp->exp_flvr;
- exp->exp_flvr_expire[0] = cfs_time_current_sec() +
+ exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
EXP_FLVR_UPDATE_EXPIRE;
exp->exp_flvr = flavor;
/* if it's gss, we only interested in root ctx init */
if (req->rq_auth_gss &&
- !(req->rq_ctx_init && (req->rq_auth_usr_root ||
- req->rq_auth_usr_mdt))) {
- spin_unlock(&exp->exp_lock);
- CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d)\n",
- req->rq_auth_gss, req->rq_ctx_init,
- req->rq_auth_usr_root, req->rq_auth_usr_mdt);
- return 0;
- }
-
- exp->exp_flvr_adapt = 0;
- spin_unlock(&exp->exp_lock);
+ !(req->rq_ctx_init &&
+ (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
+ req->rq_auth_usr_ost))) {
+ spin_unlock(&exp->exp_lock);
+ CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
+ req->rq_auth_gss, req->rq_ctx_init,
+ req->rq_auth_usr_root, req->rq_auth_usr_mdt,
+ req->rq_auth_usr_ost);
+ return 0;
+ }
+
+ exp->exp_flvr_adapt = 0;
+ spin_unlock(&exp->exp_lock);
return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
- req->rq_svc_ctx, flavor.sf_rpc);
+ req->rq_svc_ctx, &flavor);
}
/* if it equals to the current flavor, we accept it, but need to
/* most cases should return here, we only interested in
* gss root ctx init */
if (!req->rq_auth_gss || !req->rq_ctx_init ||
- (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt)) {
- spin_unlock(&exp->exp_lock);
- return 0;
- }
-
- /* if flavor just changed, we should not proceed, just leave
- * it and current flavor will be discovered and replaced
- * shortly, and let _this_ rpc pass through */
- if (exp->exp_flvr_changed) {
- LASSERT(exp->exp_flvr_adapt);
- spin_unlock(&exp->exp_lock);
- return 0;
- }
-
- if (exp->exp_flvr_adapt) {
- exp->exp_flvr_adapt = 0;
- CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
- exp, exp->exp_flvr.sf_rpc,
- exp->exp_flvr_old[0].sf_rpc,
- exp->exp_flvr_old[1].sf_rpc);
- flavor = exp->exp_flvr;
- spin_unlock(&exp->exp_lock);
-
- return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
- req->rq_svc_ctx,
- flavor.sf_rpc);
- } else {
- CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
- "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
- exp->exp_flvr_old[0].sf_rpc,
- exp->exp_flvr_old[1].sf_rpc);
- spin_unlock(&exp->exp_lock);
+ (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
+ !req->rq_auth_usr_ost)) {
+ spin_unlock(&exp->exp_lock);
+ return 0;
+ }
+
+ /* if flavor just changed, we should not proceed, just leave
+ * it and current flavor will be discovered and replaced
+ * shortly, and let _this_ rpc pass through */
+ if (exp->exp_flvr_changed) {
+ LASSERT(exp->exp_flvr_adapt);
+ spin_unlock(&exp->exp_lock);
+ return 0;
+ }
+
+ if (exp->exp_flvr_adapt) {
+ exp->exp_flvr_adapt = 0;
+ CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
+ exp, exp->exp_flvr.sf_rpc,
+ exp->exp_flvr_old[0].sf_rpc,
+ exp->exp_flvr_old[1].sf_rpc);
+ flavor = exp->exp_flvr;
+ spin_unlock(&exp->exp_lock);
+
+ return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
+ req->rq_svc_ctx,
+ &flavor);
+ } else {
+ CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
+ "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
+ exp->exp_flvr_old[0].sf_rpc,
+ exp->exp_flvr_old[1].sf_rpc);
+ spin_unlock(&exp->exp_lock);
return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
req->rq_svc_ctx);
}
if (exp->exp_flvr_expire[0]) {
- if (exp->exp_flvr_expire[0] >= cfs_time_current_sec()) {
+ if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
- CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
- "middle one ("CFS_DURATION_T")\n", exp,
+ CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the middle one (%lld)\n", exp,
exp->exp_flvr.sf_rpc,
exp->exp_flvr_old[0].sf_rpc,
exp->exp_flvr_old[1].sf_rpc,
- exp->exp_flvr_expire[0] -
- cfs_time_current_sec());
- spin_unlock(&exp->exp_lock);
+ (s64)(exp->exp_flvr_expire[0] -
+ ktime_get_real_seconds()));
+ spin_unlock(&exp->exp_lock);
return 0;
}
} else {
/* now it doesn't match the current flavor, the only chance we can
* accept it is match the old flavors which is not expired. */
if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
- if (exp->exp_flvr_expire[1] >= cfs_time_current_sec()) {
+ if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
- CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the "
- "oldest one ("CFS_DURATION_T")\n", exp,
+ CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
+ exp,
exp->exp_flvr.sf_rpc,
exp->exp_flvr_old[0].sf_rpc,
exp->exp_flvr_old[1].sf_rpc,
- exp->exp_flvr_expire[1] -
- cfs_time_current_sec());
- spin_unlock(&exp->exp_lock);
+ (s64)(exp->exp_flvr_expire[1] -
+ ktime_get_real_seconds()));
+ spin_unlock(&exp->exp_lock);
return 0;
}
} else {
exp->exp_flvr_old[1].sf_rpc);
}
- spin_unlock(&exp->exp_lock);
+ spin_unlock(&exp->exp_lock);
- CWARN("req %p: (%u|%u|%u|%u|%u) with unauthorized flavor %x\n",
+ CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
+ exp, exp->exp_obd->obd_name,
req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
- req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_flvr.sf_rpc);
+ req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
+ req->rq_flvr.sf_rpc,
+ exp->exp_flvr.sf_rpc,
+ exp->exp_flvr_old[0].sf_rpc,
+ exp->exp_flvr_expire[0] ?
+ (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
+ exp->exp_flvr_old[1].sf_rpc,
+ exp->exp_flvr_expire[1] ?
+ (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
return -EACCES;
}
+EXPORT_SYMBOL(sptlrpc_target_export_check);
void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
struct sptlrpc_rule_set *rset)
LASSERT(obd);
- spin_lock(&obd->obd_dev_lock);
+ spin_lock(&obd->obd_dev_lock);
- list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
- if (exp->exp_connection == NULL)
- continue;
+ list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
+ if (exp->exp_connection == NULL)
+ continue;
- /* note if this export had just been updated flavor
- * (exp_flvr_changed == 1), this will override the
- * previous one. */
- spin_lock(&exp->exp_lock);
- sptlrpc_rule_set_choose(rset, exp->exp_sp_peer,
- exp->exp_connection->c_peer.nid,
- &new_flvr);
+ /* note if this export had just been updated flavor
+ * (exp_flvr_changed == 1), this will override the
+ * previous one. */
+ spin_lock(&exp->exp_lock);
+ sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
+ exp->exp_connection->c_peer.nid,
+ &new_flvr);
if (exp->exp_flvr_changed ||
- memcmp(&new_flvr, &exp->exp_flvr, sizeof(new_flvr))) {
+ !flavor_equal(&new_flvr, &exp->exp_flvr)) {
exp->exp_flvr_old[1] = new_flvr;
exp->exp_flvr_expire[1] = 0;
exp->exp_flvr_changed = 1;
exp->exp_flvr_adapt = 1;
+
CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
exp, sptlrpc_part2name(exp->exp_sp_peer),
exp->exp_flvr.sf_rpc,
exp->exp_flvr_old[1].sf_rpc);
}
- spin_unlock(&exp->exp_lock);
- }
+ spin_unlock(&exp->exp_lock);
+ }
- spin_unlock(&obd->obd_dev_lock);
+ spin_unlock(&obd->obd_dev_lock);
}
EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
{
- if (svc_rc == SECSVC_DROP)
- return SECSVC_DROP;
+ /* peer's claim is unreliable unless gss is being used */
+ if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
+ return svc_rc;
switch (req->rq_sp_from) {
case LUSTRE_SP_CLI:
+ if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
+ DEBUG_REQ(D_ERROR, req, "faked source CLI");
+ svc_rc = SECSVC_DROP;
+ }
+ break;
case LUSTRE_SP_MDT:
+ if (!req->rq_auth_usr_mdt) {
+ DEBUG_REQ(D_ERROR, req, "faked source MDT");
+ svc_rc = SECSVC_DROP;
+ }
+ break;
case LUSTRE_SP_OST:
+ if (!req->rq_auth_usr_ost) {
+ DEBUG_REQ(D_ERROR, req, "faked source OST");
+ svc_rc = SECSVC_DROP;
+ }
+ break;
case LUSTRE_SP_MGS:
- case LUSTRE_SP_ANY:
+ case LUSTRE_SP_MGC:
+ if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
+ !req->rq_auth_usr_ost) {
+ DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
+ svc_rc = SECSVC_DROP;
+ }
break;
+ case LUSTRE_SP_ANY:
default:
DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
- return SECSVC_DROP;
- }
-
- if (!req->rq_auth_gss)
- return svc_rc;
-
- if (unlikely(req->rq_sp_from == LUSTRE_SP_ANY)) {
- CERROR("not specific part\n");
- return SECSVC_DROP;
- }
-
- /* from MDT, must be authenticated as MDT */
- if (unlikely(req->rq_sp_from == LUSTRE_SP_MDT &&
- !req->rq_auth_usr_mdt)) {
- DEBUG_REQ(D_ERROR, req, "fake source MDT");
- return SECSVC_DROP;
- }
-
- /* from OST, must be callback to MDT and CLI, the reverse sec
- * was from mdt/root keytab, so it should be MDT or root FIXME */
- if (unlikely(req->rq_sp_from == LUSTRE_SP_OST &&
- !req->rq_auth_usr_mdt && !req->rq_auth_usr_root)) {
- DEBUG_REQ(D_ERROR, req, "fake source OST");
- return SECSVC_DROP;
+ svc_rc = SECSVC_DROP;
}
return svc_rc;
}
+/**
+ * Used by ptlrpc server, to perform transformation upon request message of
+ * incoming \a req. This must be the first thing to do with an incoming
+ * request in ptlrpc layer.
+ *
+ * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
+ * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
+ * \retval SECSVC_COMPLETE success, the request has been fully processed, and
+ * reply message has been prepared.
+ * \retval SECSVC_DROP failed, this request should be dropped.
+ */
int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
{
struct ptlrpc_sec_policy *policy;
- struct lustre_msg *msg = req->rq_reqbuf;
- int rc;
+ struct lustre_msg *msg = req->rq_reqbuf;
+ int rc;
ENTRY;
LASSERT(msg);
LASSERT(req->rq_reqmsg == NULL);
LASSERT(req->rq_repmsg == NULL);
+ LASSERT(req->rq_svc_ctx == NULL);
- req->rq_sp_from = LUSTRE_SP_ANY;
- req->rq_auth_uid = INVALID_UID;
- req->rq_auth_mapped_uid = INVALID_UID;
+ req->rq_req_swab_mask = 0;
- if (req->rq_reqdata_len < sizeof(struct lustre_msg)) {
- CERROR("request size %d too small\n", req->rq_reqdata_len);
+ rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
+ switch (rc) {
+ case 1:
+ lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
+ case 0:
+ break;
+ default:
+ CERROR("error unpacking request from %s x%llu\n",
+ libcfs_id2str(req->rq_peer), req->rq_xid);
RETURN(SECSVC_DROP);
}
- /*
- * v2 message.
- */
- if (msg->lm_magic == LUSTRE_MSG_MAGIC_V2)
- req->rq_flvr.sf_rpc = WIRE_FLVR_RPC(msg->lm_secflvr);
- else
- req->rq_flvr.sf_rpc = WIRE_FLVR_RPC(__swab32(msg->lm_secflvr));
-
- /* unpack the wrapper message if the policy is not null */
- if ((RPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL) &&
- lustre_unpack_msg(msg, req->rq_reqdata_len))
- RETURN(SECSVC_DROP);
+ req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
+ req->rq_sp_from = LUSTRE_SP_ANY;
+ req->rq_auth_uid = -1; /* set to INVALID_UID */
+ req->rq_auth_mapped_uid = -1;
- policy = sptlrpc_rpcflavor2policy(req->rq_flvr.sf_rpc);
+ policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
if (!policy) {
CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
RETURN(SECSVC_DROP);
LASSERT(policy->sp_sops->accept);
rc = policy->sp_sops->accept(req);
-
- LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
sptlrpc_policy_put(policy);
+ LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
+ LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
+
+ /*
+ * if it's not null flavor (which means embedded packing msg),
+ * reset the swab mask for the comming inner msg unpacking.
+ */
+ if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
+ req->rq_req_swab_mask = 0;
/* sanity check for the request source */
rc = sptlrpc_svc_check_from(req, rc);
-
- /* FIXME move to proper place */
- if (rc == SECSVC_OK) {
- __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
-
- if (opc == OST_WRITE)
- req->rq_bulk_write = 1;
- else if (opc == OST_READ)
- req->rq_bulk_read = 1;
- }
-
- LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
RETURN(rc);
}
-int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req,
- int msglen)
+/**
+ * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
+ * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
+ * a buffer of \a msglen size.
+ */
+int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
{
struct ptlrpc_sec_policy *policy;
struct ptlrpc_reply_state *rs;
rc = policy->sp_sops->alloc_rs(req, msglen);
if (unlikely(rc == -ENOMEM)) {
+ struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
+ if (svcpt->scp_service->srv_max_reply_size <
+ msglen + sizeof(struct ptlrpc_reply_state)) {
+ /* Just return failure if the size is too big */
+ CERROR("size of message is too big (%zd), %d allowed\n",
+ msglen + sizeof(struct ptlrpc_reply_state),
+ svcpt->scp_service->srv_max_reply_size);
+ RETURN(-ENOMEM);
+ }
+
/* failed alloc, try emergency pool */
- rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service);
+ rs = lustre_get_emerg_rs(svcpt);
if (rs == NULL)
RETURN(-ENOMEM);
RETURN(rc);
}
+/**
+ * Used by ptlrpc server, to perform transformation upon reply message.
+ *
+ * \post req->rq_reply_off is set to approriate server-controlled reply offset.
+ * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
+ */
int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
{
struct ptlrpc_sec_policy *policy;
RETURN(rc);
}
+/**
+ * Used by ptlrpc server, to free reply_state.
+ */
void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
{
struct ptlrpc_sec_policy *policy;
void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
{
- struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
-
- if (ctx == NULL)
- return;
+ struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
- LASSERT(atomic_read(&ctx->sc_refcount) > 0);
- atomic_inc(&ctx->sc_refcount);
+ if (ctx != NULL)
+ atomic_inc(&ctx->sc_refcount);
}
void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
{
- struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
+ struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
- if (ctx == NULL)
- return;
+ if (ctx == NULL)
+ return;
- LASSERT(atomic_read(&ctx->sc_refcount) > 0);
- if (atomic_dec_and_test(&ctx->sc_refcount)) {
- if (ctx->sc_policy->sp_sops->free_ctx)
- ctx->sc_policy->sp_sops->free_ctx(ctx);
- }
- req->rq_svc_ctx = NULL;
+ LASSERT_ATOMIC_POS(&ctx->sc_refcount);
+ if (atomic_dec_and_test(&ctx->sc_refcount)) {
+ if (ctx->sc_policy->sp_sops->free_ctx)
+ ctx->sc_policy->sp_sops->free_ctx(ctx);
+ }
+ req->rq_svc_ctx = NULL;
}
void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
if (ctx == NULL)
return;
- LASSERT(atomic_read(&ctx->sc_refcount) > 0);
+ LASSERT_ATOMIC_POS(&ctx->sc_refcount);
if (ctx->sc_policy->sp_sops->invalidate_ctx)
ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
}
* bulk security *
****************************************/
+/**
+ * Perform transformation upon bulk data pointed by \a desc. This is called
+ * before transforming the request message.
+ */
int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
struct ptlrpc_bulk_desc *desc)
{
struct ptlrpc_cli_ctx *ctx;
+ LASSERT(req->rq_bulk_read || req->rq_bulk_write);
+
if (!req->rq_pack_bulk)
return 0;
- LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
ctx = req->rq_cli_ctx;
if (ctx->cc_ops->wrap_bulk)
return ctx->cc_ops->wrap_bulk(ctx, req, desc);
}
EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
-static
-void pga_to_bulk_desc(int nob, obd_count pg_count, struct brw_page **pga,
- struct ptlrpc_bulk_desc *desc)
-{
- int i;
-
- LASSERT(pga);
- LASSERT(*pga);
-
- for (i = 0; i < pg_count && nob > 0; i++) {
-#ifdef __KERNEL__
- desc->bd_iov[i].kiov_page = pga[i]->pg;
- desc->bd_iov[i].kiov_len = pga[i]->count > nob ?
- nob : pga[i]->count;
- desc->bd_iov[i].kiov_offset = pga[i]->off & ~CFS_PAGE_MASK;
-#else
- /* FIXME currently liblustre doesn't support bulk encryption.
- * if we do, check again following may not be right. */
- LASSERTF(0, "Bulk encryption not implemented for liblustre\n");
- desc->bd_iov[i].iov_base = pga[i]->pg->addr;
- desc->bd_iov[i].iov_len = pga[i]->count > nob ?
- nob : pga[i]->count;
-#endif
-
- desc->bd_iov_count++;
- nob -= pga[i]->count;
- }
-}
-
+/**
+ * This is called after unwrap the reply message.
+ * return nob of actual plain text size received, or error code.
+ */
int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
- int nob, obd_count pg_count,
- struct brw_page **pga)
+ struct ptlrpc_bulk_desc *desc,
+ int nob)
{
- struct ptlrpc_bulk_desc *desc;
- struct ptlrpc_cli_ctx *ctx;
- int rc = 0;
-
- if (!req->rq_pack_bulk)
- return 0;
+ struct ptlrpc_cli_ctx *ctx;
+ int rc;
LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
- OBD_ALLOC(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
- if (desc == NULL) {
- CERROR("out of memory, can't verify bulk read data\n");
- return -ENOMEM;
- }
-
- pga_to_bulk_desc(nob, pg_count, pga, desc);
+ if (!req->rq_pack_bulk)
+ return desc->bd_nob_transferred;
ctx = req->rq_cli_ctx;
- if (ctx->cc_ops->unwrap_bulk)
+ if (ctx->cc_ops->unwrap_bulk) {
rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
-
- OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
-
- return rc;
+ if (rc < 0)
+ return rc;
+ }
+ return desc->bd_nob_transferred;
}
EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
+/**
+ * This is called after unwrap the reply message.
+ * return 0 for success or error code.
+ */
int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
struct ptlrpc_bulk_desc *desc)
{
- struct ptlrpc_cli_ctx *ctx;
+ struct ptlrpc_cli_ctx *ctx;
+ int rc;
+
+ LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
if (!req->rq_pack_bulk)
return 0;
- LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
-
ctx = req->rq_cli_ctx;
- if (ctx->cc_ops->unwrap_bulk)
- return ctx->cc_ops->unwrap_bulk(ctx, req, desc);
+ if (ctx->cc_ops->unwrap_bulk) {
+ rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
+ if (rc < 0)
+ return rc;
+ }
- return 0;
+ /*
+ * if everything is going right, nob should equals to nob_transferred.
+ * in case of privacy mode, nob_transferred needs to be adjusted.
+ */
+ if (desc->bd_nob != desc->bd_nob_transferred) {
+ CERROR("nob %d doesn't match transferred nob %d\n",
+ desc->bd_nob, desc->bd_nob_transferred);
+ return -EPROTO;
+ }
+
+ return 0;
}
EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
+#ifdef HAVE_SERVER_SUPPORT
+/**
+ * Performe transformation upon outgoing bulk read.
+ */
int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
struct ptlrpc_bulk_desc *desc)
{
struct ptlrpc_svc_ctx *ctx;
+ LASSERT(req->rq_bulk_read);
+
if (!req->rq_pack_bulk)
return 0;
- LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
ctx = req->rq_svc_ctx;
if (ctx->sc_policy->sp_sops->wrap_bulk)
return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
}
EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
+/**
+ * Performe transformation upon incoming bulk write.
+ */
int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
struct ptlrpc_bulk_desc *desc)
{
struct ptlrpc_svc_ctx *ctx;
+ int rc;
+
+ LASSERT(req->rq_bulk_write);
+
+ /*
+ * if it's in privacy mode, transferred should >= expected; otherwise
+ * transferred should == expected.
+ */
+ if (desc->bd_nob_transferred < desc->bd_nob ||
+ (desc->bd_nob_transferred > desc->bd_nob &&
+ SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
+ SPTLRPC_BULK_SVC_PRIV)) {
+ DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
+ desc->bd_nob_transferred, desc->bd_nob);
+ return -ETIMEDOUT;
+ }
if (!req->rq_pack_bulk)
return 0;
- LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
ctx = req->rq_svc_ctx;
- if (ctx->sc_policy->sp_sops->unwrap_bulk);
- return ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
+ if (ctx->sc_policy->sp_sops->unwrap_bulk) {
+ rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
+ if (rc)
+ CERROR("error unwrap bulk: %d\n", rc);
+ }
+ /* return 0 to allow reply be sent */
return 0;
}
EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
+/**
+ * Prepare buffers for incoming bulk write.
+ */
+int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
+ struct ptlrpc_bulk_desc *desc)
+{
+ struct ptlrpc_svc_ctx *ctx;
+
+ LASSERT(req->rq_bulk_write);
+
+ if (!req->rq_pack_bulk)
+ return 0;
+
+ ctx = req->rq_svc_ctx;
+ if (ctx->sc_policy->sp_sops->prep_bulk)
+ return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
+
+ return 0;
+}
+EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
+
+#endif /* HAVE_SERVER_SUPPORT */
/****************************************
* user descriptor helpers *
{
int ngroups;
-#ifdef __KERNEL__
ngroups = current_ngroups;
if (ngroups > LUSTRE_MAX_GROUPS)
ngroups = LUSTRE_MAX_GROUPS;
-#else
- ngroups = 0;
-#endif
return sptlrpc_user_desc_size(ngroups);
}
EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
{
- struct ptlrpc_user_desc *pud;
+ struct ptlrpc_user_desc *pud;
- pud = lustre_msg_buf(msg, offset, 0);
-
- pud->pud_uid = cfs_current()->uid;
- pud->pud_gid = cfs_current()->gid;
- pud->pud_fsuid = cfs_current()->fsuid;
- pud->pud_fsgid = cfs_current()->fsgid;
- pud->pud_cap = cfs_curproc_cap_pack();
- pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
-
-#ifdef __KERNEL__
- task_lock(current);
- if (pud->pud_ngroups > current_ngroups)
- pud->pud_ngroups = current_ngroups;
- memcpy(pud->pud_groups, cfs_current()->group_info->blocks[0],
- pud->pud_ngroups * sizeof(__u32));
- task_unlock(current);
-#endif
+ pud = lustre_msg_buf(msg, offset, 0);
- return 0;
+ pud->pud_uid = from_kuid(&init_user_ns, current_uid());
+ pud->pud_gid = from_kgid(&init_user_ns, current_gid());
+ pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
+ pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
+ pud->pud_cap = cfs_curproc_cap_pack();
+ pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
+
+ task_lock(current);
+ if (pud->pud_ngroups > current_ngroups)
+ pud->pud_ngroups = current_ngroups;
+#ifdef HAVE_GROUP_INFO_GID
+ memcpy(pud->pud_groups, current_cred()->group_info->gid,
+ pud->pud_ngroups * sizeof(__u32));
+#else /* !HAVE_GROUP_INFO_GID */
+ memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
+ pud->pud_ngroups * sizeof(__u32));
+#endif /* HAVE_GROUP_INFO_GID */
+ task_unlock(current);
+
+ return 0;
}
EXPORT_SYMBOL(sptlrpc_pack_user_desc);
-int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset)
+int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
{
struct ptlrpc_user_desc *pud;
int i;
if (!pud)
return -EINVAL;
- if (lustre_msg_swabbed(msg)) {
+ if (swabbed) {
__swab32s(&pud->pud_uid);
__swab32s(&pud->pud_gid);
__swab32s(&pud->pud_fsuid);
return -EINVAL;
}
- if (lustre_msg_swabbed(msg)) {
+ if (swabbed) {
for (i = 0; i < pud->pud_ngroups; i++)
__swab32s(&pud->pud_groups[i]);
}
}
EXPORT_SYMBOL(sec2target_str);
+/*
+ * return true if the bulk data is protected
+ */
+int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
+{
+ switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
+ case SPTLRPC_BULK_SVC_INTG:
+ case SPTLRPC_BULK_SVC_PRIV:
+ return 1;
+ default:
+ return 0;
+ }
+}
+EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
+
/****************************************
* crypto API helper/alloc blkciper *
****************************************/
-#ifdef __KERNEL__
-#ifndef HAVE_ASYNC_BLOCK_CIPHER
-struct ll_crypto_cipher *ll_crypto_alloc_blkcipher(const char * algname,
- u32 type, u32 mask)
-{
- char buf[CRYPTO_MAX_ALG_NAME + 1];
- const char *pan = algname;
- u32 flag = 0;
-
- if (strncmp("cbc(", algname, 4) == 0)
- flag |= CRYPTO_TFM_MODE_CBC;
- else if (strncmp("ecb(", algname, 4) == 0)
- flag |= CRYPTO_TFM_MODE_ECB;
- if (flag) {
- char *vp = strnchr(algname, CRYPTO_MAX_ALG_NAME, ')');
- if (vp) {
- memcpy(buf, algname + 4, vp - algname - 4);
- buf[vp - algname - 4] = '\0';
- pan = buf;
- } else {
- flag = 0;
- }
- }
- return crypto_alloc_tfm(pan, flag);
-}
-EXPORT_SYMBOL(ll_crypto_alloc_blkcipher);
-#endif
-#endif
-
/****************************************
* initialize/finalize *
****************************************/
-int __init sptlrpc_init(void)
+int sptlrpc_init(void)
{
int rc;
- rc = sptlrpc_gc_start_thread();
+ rwlock_init(&policy_lock);
+
+ rc = sptlrpc_gc_init();
if (rc)
goto out;
- rc = sptlrpc_enc_pool_init();
+ rc = sptlrpc_conf_init();
if (rc)
goto out_gc;
+ rc = sptlrpc_enc_pool_init();
+ if (rc)
+ goto out_conf;
+
rc = sptlrpc_null_init();
if (rc)
goto out_pool;
sptlrpc_null_fini();
out_pool:
sptlrpc_enc_pool_fini();
+out_conf:
+ sptlrpc_conf_fini();
out_gc:
- sptlrpc_gc_stop_thread();
+ sptlrpc_gc_fini();
out:
return rc;
}
-void __exit sptlrpc_fini(void)
+void sptlrpc_fini(void)
{
sptlrpc_lproc_fini();
sptlrpc_plain_fini();
sptlrpc_null_fini();
sptlrpc_enc_pool_fini();
- sptlrpc_gc_stop_thread();
+ sptlrpc_conf_fini();
+ sptlrpc_gc_fini();
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff