*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
*
* Author: Eric Mei <ericm@clusterfs.com>
*/
#include <linux/hash.h>
#include <linux/mutex.h>
#include <linux/sunrpc/cache.h>
+#include <net/sock.h>
#include <obd.h>
#include <obd_class.h>
#include <obd_support.h>
-#include <lustre/lustre_idl.h>
-#include <lustre_net.h>
#include <lustre_import.h>
+#include <lustre_net.h>
+#include <lustre_nodemap.h>
#include <lustre_sec.h>
#include "gss_err.h"
#define GSS_SVC_UPCALL_TIMEOUT (20)
-static spinlock_t __ctx_index_lock;
+static DEFINE_SPINLOCK(__ctx_index_lock);
static __u64 __ctx_index;
__u64 gss_get_next_ctx_index(void)
#endif
}
/****************************************
- * rsi cache *
+ * rpc sec init (rsi) cache *
****************************************/
#define RSI_HASHBITS (6)
struct cache_head h;
__u32 lustre_svc;
__u64 nid;
+ char nm_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
wait_queue_head_t waitq;
rawobj_t in_handle, in_token;
rawobj_t out_handle, out_token;
int major_status, minor_status;
};
+#ifdef HAVE_CACHE_HEAD_HLIST
+static struct hlist_head rsi_table[RSI_HASHMAX];
+#else
static struct cache_head *rsi_table[RSI_HASHMAX];
+#endif
static struct cache_detail rsi_cache;
static struct rsi *rsi_update(struct rsi *new, struct rsi *old);
static struct rsi *rsi_lookup(struct rsi *item);
rawobj_free(&rsi->out_token);
}
+/* See handle_channel_req() userspace for where the upcall data is read */
static void rsi_request(struct cache_detail *cd,
struct cache_head *h,
char **bpp, int *blen)
sizeof(rsi->lustre_svc));
qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid));
qword_addhex(bpp, blen, (char *) &index, sizeof(index));
+ qword_addhex(bpp, blen, (char *) rsi->nm_name,
+ strlen(rsi->nm_name) + 1);
qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len);
qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len);
(*bpp)[-1] = '\n';
new->lustre_svc = item->lustre_svc;
new->nid = item->nid;
+ memcpy(new->nm_name, item->nm_name, sizeof(item->nm_name));
init_waitqueue_head(&new->waitq);
}
{
struct rsi *rsi = container_of(ref, struct rsi, h.ref);
- LASSERT(rsi->h.next == NULL);
+#ifdef HAVE_CACHE_HEAD_HLIST
+ LASSERT(rsi->h.cache_list.next == NULL);
+#else
+ LASSERT(rsi->h.next == NULL);
+#endif
rsi_free(rsi);
OBD_FREE_PTR(rsi);
}
static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen)
{
char *buf = mesg;
- char *ep;
int len;
struct rsi rsii, *rsip = NULL;
time_t expiry;
if (len <= 0)
goto out;
- /* major */
- rsii.major_status = simple_strtol(buf, &ep, 10);
- if (*ep)
- goto out;
+ /* major */
+ status = kstrtoint(buf, 10, &rsii.major_status);
+ if (status)
+ goto out;
- /* minor */
- len = qword_get(&mesg, buf, mlen);
- if (len <= 0)
- goto out;
- rsii.minor_status = simple_strtol(buf, &ep, 10);
- if (*ep)
- goto out;
+ /* minor */
+ len = qword_get(&mesg, buf, mlen);
+ if (len <= 0) {
+ status = -EINVAL;
+ goto out;
+ }
+
+ status = kstrtoint(buf, 10, &rsii.minor_status);
+ if (status)
+ goto out;
/* out_handle */
len = qword_get(&mesg, buf, mlen);
}
/****************************************
- * rsc cache *
+ * rpc sec context (rsc) cache *
****************************************/
#define RSC_HASHBITS (10)
struct gss_svc_ctx ctx;
};
+#ifdef HAVE_CACHE_HEAD_HLIST
+static struct hlist_head rsc_table[RSC_HASHMAX];
+#else
static struct cache_head *rsc_table[RSC_HASHMAX];
+#endif
static struct cache_detail rsc_cache;
static struct rsc *rsc_update(struct rsc *new, struct rsc *old);
static struct rsc *rsc_lookup(struct rsc *item);
{
struct rsc *rsci = container_of(ref, struct rsc, h.ref);
+#ifdef HAVE_CACHE_HEAD_HLIST
+ LASSERT(rsci->h.cache_list.next == NULL);
+#else
LASSERT(rsci->h.next == NULL);
+#endif
rsc_free(rsci);
OBD_FREE_PTR(rsci);
}
}
rsci.ctx.gsc_remote = (tmp_int != 0);
- /* root user flag */
- rv = get_int(&mesg, &tmp_int);
- if (rv) {
- CERROR("fail to get oss user flag\n");
- goto out;
- }
- rsci.ctx.gsc_usr_root = (tmp_int != 0);
+ /* root user flag */
+ rv = get_int(&mesg, &tmp_int);
+ if (rv) {
+ CERROR("fail to get root user flag\n");
+ goto out;
+ }
+ rsci.ctx.gsc_usr_root = (tmp_int != 0);
/* mds user flag */
rv = get_int(&mesg, &tmp_int);
CERROR("NOENT? set rsc entry negative\n");
set_bit(CACHE_NEGATIVE, &rsci.h.flags);
} else {
- rawobj_t tmp_buf;
- unsigned long ctx_expiry;
-
- /* gid */
- if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
- goto out;
-
- /* mech name */
- len = qword_get(&mesg, buf, mlen);
- if (len < 0)
- goto out;
- gm = lgss_name_to_mech(buf);
- status = -EOPNOTSUPP;
- if (!gm)
- goto out;
-
- status = -EINVAL;
- /* mech-specific data: */
- len = qword_get(&mesg, buf, mlen);
- if (len < 0)
- goto out;
-
- tmp_buf.len = len;
- tmp_buf.data = (unsigned char *)buf;
- if (lgss_import_sec_context(&tmp_buf, gm,
- &rsci.ctx.gsc_mechctx))
- goto out;
-
- /* currently the expiry time passed down from user-space
- * is invalid, here we retrive it from mech. */
- if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
- CERROR("unable to get expire time, drop it\n");
- goto out;
- }
- expiry = (time_t) ctx_expiry;
+ rawobj_t tmp_buf;
+ time64_t ctx_expiry;
+
+ /* gid */
+ if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid))
+ goto out;
+
+ /* mech name */
+ len = qword_get(&mesg, buf, mlen);
+ if (len < 0)
+ goto out;
+ gm = lgss_name_to_mech(buf);
+ status = -EOPNOTSUPP;
+ if (!gm)
+ goto out;
+
+ status = -EINVAL;
+ /* mech-specific data: */
+ len = qword_get(&mesg, buf, mlen);
+ if (len < 0)
+ goto out;
+
+ tmp_buf.len = len;
+ tmp_buf.data = (unsigned char *)buf;
+ if (lgss_import_sec_context(&tmp_buf, gm,
+ &rsci.ctx.gsc_mechctx))
+ goto out;
+
+ /* set to seconds since machine booted */
+ expiry = ktime_get_seconds();
+
+ /* currently the expiry time passed down from user-space
+ * is invalid, here we retrive it from mech.
+ */
+ if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) {
+ CERROR("unable to get expire time, drop it\n");
+ goto out;
+ }
+
+ /* ctx_expiry is the number of seconds since Jan 1 1970.
+ * We want just the number of seconds into the future.
+ */
+ expiry += ctx_expiry - ktime_get_real_seconds();
}
rsci.h.expiry_time = expiry;
static void rsc_flush(rsc_entry_match *match, long data)
{
- struct cache_head **ch;
+#ifdef HAVE_CACHE_HEAD_HLIST
+ struct cache_head *ch = NULL;
+ struct hlist_head *head;
+#else
+ struct cache_head **ch;
+#endif
struct rsc *rscp;
int n;
ENTRY;
write_lock(&rsc_cache.hash_lock);
for (n = 0; n < RSC_HASHMAX; n++) {
- for (ch = &rsc_cache.hash_table[n]; *ch;) {
- rscp = container_of(*ch, struct rsc, h);
+#ifdef HAVE_CACHE_HEAD_HLIST
+ head = &rsc_cache.hash_table[n];
+ hlist_for_each_entry(ch, head, cache_list) {
+ rscp = container_of(ch, struct rsc, h);
+#else
+ for (ch = &rsc_cache.hash_table[n]; *ch;) {
+ rscp = container_of(*ch, struct rsc, h);
+#endif
if (!match(rscp, data)) {
- ch = &((*ch)->next);
+#ifndef HAVE_CACHE_HEAD_HLIST
+ ch = &((*ch)->next);
+#endif
continue;
}
/* it seems simply set NEGATIVE doesn't work */
- *ch = (*ch)->next;
- rscp->h.next = NULL;
+#ifdef HAVE_CACHE_HEAD_HLIST
+ hlist_del_init(&ch->cache_list);
+#else
+ *ch = (*ch)->next;
+ rscp->h.next = NULL;
+#endif
cache_get(&rscp->h);
set_bit(CACHE_NEGATIVE, &rscp->h.flags);
COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
{
rsc_flush_target(target);
}
-EXPORT_SYMBOL(gss_secsvc_flush);
static struct rsc *gss_svc_searchbyctx(rawobj_t *handle)
{
struct gss_cli_ctx *gctx)
{
struct rsc rsci, *rscp = NULL;
- unsigned long ctx_expiry;
+ time64_t ctx_expiry;
__u32 major;
int rc;
ENTRY;
}
rsci.h.expiry_time = (time_t) ctx_expiry;
- if (strcmp(imp->imp_obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0)
- rsci.ctx.gsc_usr_mds = 1;
- else if (strcmp(imp->imp_obd->obd_type->typ_name, LUSTRE_OSC_NAME) == 0)
- rsci.ctx.gsc_usr_oss = 1;
- else
- rsci.ctx.gsc_usr_root = 1;
+ switch (imp->imp_obd->u.cli.cl_sp_to) {
+ case LUSTRE_SP_MDT:
+ rsci.ctx.gsc_usr_mds = 1;
+ break;
+ case LUSTRE_SP_OST:
+ rsci.ctx.gsc_usr_oss = 1;
+ break;
+ case LUSTRE_SP_CLI:
+ rsci.ctx.gsc_usr_root = 1;
+ break;
+ case LUSTRE_SP_MGS:
+ /* by convention, all 3 set to 1 means MGS */
+ rsci.ctx.gsc_usr_mds = 1;
+ rsci.ctx.gsc_usr_oss = 1;
+ rsci.ctx.gsc_usr_root = 1;
+ break;
+ default:
+ break;
+ }
rscp = rsc_update(&rsci, rscp);
if (rscp == NULL)
rscp->target = imp->imp_obd;
rawobj_dup(&gctx->gc_svc_handle, &rscp->handle);
- CWARN("create reverse svc ctx %p to %s: idx "LPX64"\n",
+ CWARN("create reverse svc ctx %p to %s: idx %#llx\n",
&rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl);
rc = 0;
out:
rsc_free(&rsci);
if (rc)
- CERROR("create reverse svc ctx: idx "LPX64", rc %d\n",
+ CERROR("create reverse svc ctx: idx %#llx, rc %d\n",
gsec->gs_rvs_hdl, rc);
RETURN(rc);
}
int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle)
{
- const cfs_time_t expire = 20;
- struct rsc *rscp;
+ const time64_t expire = 20;
+ struct rsc *rscp;
rscp = gss_svc_searchbyctx(handle);
if (rscp) {
CDEBUG(D_SEC, "reverse svcctx %p (rsc %p) expire soon\n",
&rscp->ctx, rscp);
- rscp->h.expiry_time = cfs_time_current_sec() + expire;
+ rscp->h.expiry_time = ktime_get_real_seconds() + expire;
COMPAT_RSC_PUT(&rscp->h, &rsc_cache);
}
return 0;
struct ptlrpc_reply_state *rs;
struct rsc *rsci = NULL;
struct rsi *rsip = NULL, rsikey;
- wait_queue_t wait;
+ wait_queue_entry_t wait;
int replen = sizeof(struct ptlrpc_body);
struct gss_rep_header *rephdr;
int first_check = 1;
int rc = SECSVC_DROP;
ENTRY;
- memset(&rsikey, 0, sizeof(rsikey));
- rsikey.lustre_svc = lustre_svc;
- rsikey.nid = (__u64) req->rq_peer.nid;
+ memset(&rsikey, 0, sizeof(rsikey));
+ rsikey.lustre_svc = lustre_svc;
+ /* In case of MR, rq_peer is not the NID from which request is received,
+ * but primary NID of peer.
+ * So we need rq_source, which contains the NID actually in use.
+ */
+ rsikey.nid = (__u64) req->rq_source.nid;
+ nodemap_test_nid(req->rq_peer.nid, rsikey.nm_name,
+ sizeof(rsikey.nm_name));
/* duplicate context handle. for INIT it always 0 */
if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) {
cache_get(&rsip->h); /* take an extra ref */
init_waitqueue_head(&rsip->waitq);
- init_waitqueue_entry_current(&wait);
+ init_waitqueue_entry(&wait, current);
add_wait_queue(&rsip->waitq, &wait);
cache_check:
switch (rc) {
case -ETIMEDOUT:
case -EAGAIN: {
- int valid;
+ int valid;
- if (first_check) {
- first_check = 0;
+ if (first_check) {
+ first_check = 0;
- read_lock(&rsi_cache.hash_lock);
+ read_lock(&rsi_cache.hash_lock);
valid = test_bit(CACHE_VALID, &rsip->h.flags);
- if (valid == 0)
- set_current_state(TASK_INTERRUPTIBLE);
- read_unlock(&rsi_cache.hash_lock);
-
if (valid == 0)
- schedule_timeout(GSS_SVC_UPCALL_TIMEOUT *
- HZ);
-
+ set_current_state(TASK_INTERRUPTIBLE);
+ read_unlock(&rsi_cache.hash_lock);
+
+ if (valid == 0) {
+ unsigned long jiffies;
+ jiffies = msecs_to_jiffies(MSEC_PER_SEC *
+ GSS_SVC_UPCALL_TIMEOUT);
+ schedule_timeout(jiffies);
+ }
cache_get(&rsip->h);
goto cache_check;
- }
- CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
- break;
- }
- case -ENOENT:
- CWARN("cache_check return ENOENT, drop\n");
- break;
- case 0:
- /* if not the first check, we have to release the extra
- * reference we just added on it. */
+ }
+ CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT);
+ break;
+ }
+ case -ENOENT:
+ CDEBUG(D_SEC, "cache_check return ENOENT, drop\n");
+ break;
+ case 0:
+ /* if not the first check, we have to release the extra
+ * reference we just added on it. */
if (!first_check)
cache_put(&rsip->h, &rsi_cache);
CDEBUG(D_SEC, "cache_check is good\n");
if (!rsci) {
CERROR("authentication failed\n");
- if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0))
- rc = SECSVC_COMPLETE;
+ /* gss mechanism returned major and minor code so we return
+ * those in error message */
+ if (!gss_pack_err_notify(req, rsip->major_status,
+ rsip->minor_status))
+ rc = SECSVC_COMPLETE;
GOTO(out, rc);
} else {
rc = SECSVC_OK;
out:
- /* it looks like here we should put rsip also, but this mess up
- * with NFS cache mgmt code... FIXME */
-#if 0
- if (rsip)
- rsi_put(&rsip->h, &rsi_cache);
-#endif
-
- if (rsci) {
- /* if anything went wrong, we don't keep the context too */
- if (rc != SECSVC_OK)
+ /* it looks like here we should put rsip also, but this mess up
+ * with NFS cache mgmt code... FIXME
+ * something like:
+ * if (rsip)
+ * rsi_put(&rsip->h, &rsi_cache); */
+
+ if (rsci) {
+ /* if anything went wrong, we don't keep the context too */
+ if (rc != SECSVC_OK)
set_bit(CACHE_NEGATIVE, &rsci->h.flags);
- else
- CDEBUG(D_SEC, "create rsc with idx "LPX64"\n",
- gss_handle_to_u64(&rsci->handle));
+ else
+ CDEBUG(D_SEC, "create rsc with idx %#llx\n",
+ gss_handle_to_u64(&rsci->handle));
- COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
- }
- RETURN(rc);
+ COMPAT_RSC_PUT(&rsci->h, &rsc_cache);
+ }
+ RETURN(rc);
}
struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req,
rsc = gss_svc_searchbyctx(&gw->gw_handle);
if (!rsc) {
- CWARN("Invalid gss ctx idx "LPX64" from %s\n",
+ CWARN("Invalid gss ctx idx %#llx from %s\n",
gss_handle_to_u64(&gw->gw_handle),
libcfs_nid2str(req->rq_peer.nid));
return NULL;
{
int i, rc;
- spin_lock_init(&__ctx_index_lock);
/*
* this helps reducing context index confliction. after server reboot,
* conflicting request from clients might be filtered out by initial
/* FIXME this looks stupid. we intend to give lsvcgssd a chance to open
* the init upcall channel, otherwise there's big chance that the first
* upcall issued before the channel be opened thus nfsv4 cache code will
- * drop the request direclty, thus lead to unnecessary recovery time.
- * here we wait at miximum 1.5 seconds. */
+ * drop the request directly, thus lead to unnecessary recovery time.
+ * Here we wait at minimum 1.5 seconds.
+ */
for (i = 0; i < 6; i++) {
if (atomic_read(&rsi_cache.readers) > 0)
break;
set_current_state(TASK_UNINTERRUPTIBLE);
- LASSERT(HZ >= 4);
- schedule_timeout(HZ / 4);
+ LASSERT(msecs_to_jiffies(MSEC_PER_SEC / 4) > 0);
+ schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC / 4));
}
if (atomic_read(&rsi_cache.readers) == 0)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff