diff options
Diffstat (limited to 'rpc/rpc-lib/src')
26 files changed, 10935 insertions, 0 deletions
diff --git a/rpc/rpc-lib/src/Makefile.am b/rpc/rpc-lib/src/Makefile.am new file mode 100644 index 00000000000..35c9db07e7f --- /dev/null +++ b/rpc/rpc-lib/src/Makefile.am @@ -0,0 +1,29 @@ +lib_LTLIBRARIES = libgfrpc.la + +libgfrpc_la_SOURCES = auth-unix.c rpcsvc-auth.c rpcsvc.c auth-null.c \ + rpc-transport.c xdr-rpc.c xdr-rpcclnt.c rpc-clnt.c auth-glusterfs.c \ + rpc-drc.c rpc-clnt-ping.c \ + autoscale-threads.c mgmt-pmap.c + +EXTRA_DIST = libgfrpc.sym + +libgfrpc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \ + $(top_builddir)/rpc/xdr/src/libgfxdr.la +libgfrpc_la_LDFLAGS = -version-info $(LIBGFRPC_LT_VERSION) $(GF_LDFLAGS) \ + -export-symbols $(top_srcdir)/rpc/rpc-lib/src/libgfrpc.sym + +libgfrpc_la_HEADERS = rpcsvc.h rpc-transport.h xdr-common.h xdr-rpc.h xdr-rpcclnt.h \ + rpc-clnt.h rpcsvc-common.h protocol-common.h rpc-drc.h rpc-clnt-ping.h \ + rpc-lib-messages.h + +libgfrpc_ladir = $(includedir)/glusterfs/rpc + +AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ + -I$(top_srcdir)/rpc/xdr/src \ + -I$(top_builddir)/rpc/xdr/src \ + -DRPC_TRANSPORTDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/rpc-transport\" \ + -I$(top_srcdir)/contrib/rbtree + +AM_CFLAGS = -Wall $(GF_CFLAGS) + +CLEANFILES = *~ diff --git a/rpc/rpc-lib/src/auth-glusterfs.c b/rpc/rpc-lib/src/auth-glusterfs.c new file mode 100644 index 00000000000..69a96f7512f --- /dev/null +++ b/rpc/rpc-lib/src/auth-glusterfs.c @@ -0,0 +1,386 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#include <glusterfs/dict.h> +#include "xdr-rpc.h" +#include "xdr-common.h" +#include "rpc-common-xdr.h" +#include "glusterfs4-xdr.h" + +/* V1 */ + +ssize_t +xdr_to_glusterfs_auth(char *buf, struct auth_glusterfs_parms *req) +{ + XDR xdr; + ssize_t ret = -1; + + if ((!buf) || (!req)) + return -1; + + xdrmem_create(&xdr, buf, sizeof(struct auth_glusterfs_parms), XDR_DECODE); + if (!xdr_auth_glusterfs_parms(&xdr, req)) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs parameters"); + ret = -1; + goto ret; + } + + ret = (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)); +ret: + return ret; +} +int +auth_glusterfs_request_init(rpcsvc_request_t *req, void *priv) +{ + return 0; +} + +int +auth_glusterfs_authenticate(rpcsvc_request_t *req, void *priv) +{ + struct auth_glusterfs_parms au = { + 0, + }; + + int ret = RPCSVC_AUTH_REJECT; + int j = 0; + int i = 0; + int gidcount = 0; + + if (!req) + return ret; + + ret = xdr_to_glusterfs_auth(req->cred.authdata, &au); + if (ret == -1) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs credentials"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + req->pid = au.pid; + req->uid = au.uid; + req->gid = au.gid; + req->lk_owner.len = 8; + { + for (i = 0; i < req->lk_owner.len; i++, j += 8) + req->lk_owner.data[i] = (char)((au.lk_owner >> j) & 0xff); + } + req->auxgidcount = au.ngrps; + + if (req->auxgidcount > 16) { + gf_log("", GF_LOG_WARNING, + "more than 16 aux gids found, failing authentication"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + if (req->auxgidcount > SMALL_GROUP_COUNT) { + req->auxgidlarge = GF_CALLOC(req->auxgidcount, sizeof(req->auxgids[0]), + gf_common_mt_auxgids); + req->auxgids = req->auxgidlarge; + } else { + req->auxgids = req->auxgidsmall; + } + + if (!req->auxgids) { + gf_log("auth-glusterfs", GF_LOG_WARNING, "cannot allocate gid list"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + for (gidcount = 0; gidcount < au.ngrps; ++gidcount) + req->auxgids[gidcount] = au.groups[gidcount]; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "Auth Info: pid: %u, uid: %d" + ", gid: %d, owner: %s", + req->pid, req->uid, req->gid, lkowner_utoa(&req->lk_owner)); + ret = RPCSVC_AUTH_ACCEPT; +err: + return ret; +} + +rpcsvc_auth_ops_t auth_glusterfs_ops = { + .transport_init = NULL, + .request_init = auth_glusterfs_request_init, + .authenticate = auth_glusterfs_authenticate}; + +rpcsvc_auth_t rpcsvc_auth_glusterfs = {.authname = "AUTH_GLUSTERFS", + .authnum = AUTH_GLUSTERFS, + .authops = &auth_glusterfs_ops, + .authprivate = NULL}; + +rpcsvc_auth_t * +rpcsvc_auth_glusterfs_init(rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_glusterfs; +} + +/* V2 */ + +ssize_t +xdr_to_glusterfs_auth_v2(char *buf, struct auth_glusterfs_parms_v2 *req) +{ + XDR xdr; + ssize_t ret = -1; + + if ((!buf) || (!req)) + return -1; + + xdrmem_create(&xdr, buf, GF_MAX_AUTH_BYTES, XDR_DECODE); + if (!xdr_auth_glusterfs_parms_v2(&xdr, req)) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs v2 parameters"); + ret = -1; + goto ret; + } + + ret = (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)); +ret: + return ret; +} +int +auth_glusterfs_v2_request_init(rpcsvc_request_t *req, void *priv) +{ + return 0; +} + +int +auth_glusterfs_v2_authenticate(rpcsvc_request_t *req, void *priv) +{ + struct auth_glusterfs_parms_v2 au = { + 0, + }; + int ret = RPCSVC_AUTH_REJECT; + int i = 0; + int max_groups = 0; + int max_lk_owner_len = 0; + + if (!req) + return ret; + + ret = xdr_to_glusterfs_auth_v2(req->cred.authdata, &au); + if (ret == -1) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs credentials"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + req->pid = au.pid; + req->uid = au.uid; + req->gid = au.gid; + req->lk_owner.len = au.lk_owner.lk_owner_len; + req->auxgidcount = au.groups.groups_len; + + /* the number of groups and size of lk_owner depend on each other */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS(req->lk_owner.len, + AUTH_GLUSTERFS_v2); + max_lk_owner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER(req->auxgidcount, + AUTH_GLUSTERFS_v2); + + if (req->auxgidcount > max_groups) { + gf_log("", GF_LOG_WARNING, + "more than max aux gids found (%d) , truncating it " + "to %d and continuing", + au.groups.groups_len, max_groups); + req->auxgidcount = max_groups; + } + + if (req->lk_owner.len > max_lk_owner_len) { + gf_log("", GF_LOG_WARNING, + "lkowner field to big (%d), depends on the number of " + "groups (%d), failing authentication", + req->lk_owner.len, req->auxgidcount); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + if (req->auxgidcount > SMALL_GROUP_COUNT) { + req->auxgidlarge = GF_CALLOC(req->auxgidcount, sizeof(req->auxgids[0]), + gf_common_mt_auxgids); + req->auxgids = req->auxgidlarge; + } else { + req->auxgids = req->auxgidsmall; + } + + if (!req->auxgids) { + gf_log("auth-glusterfs-v2", GF_LOG_WARNING, "cannot allocate gid list"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + for (i = 0; i < req->auxgidcount; ++i) + req->auxgids[i] = au.groups.groups_val[i]; + + for (i = 0; i < au.lk_owner.lk_owner_len; ++i) + req->lk_owner.data[i] = au.lk_owner.lk_owner_val[i]; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "Auth Info: pid: %u, uid: %d" + ", gid: %d, owner: %s", + req->pid, req->uid, req->gid, lkowner_utoa(&req->lk_owner)); + ret = RPCSVC_AUTH_ACCEPT; +err: + /* TODO: instead use alloca() for these variables */ + free(au.groups.groups_val); + free(au.lk_owner.lk_owner_val); + + return ret; +} + +rpcsvc_auth_ops_t auth_glusterfs_ops_v2 = { + .transport_init = NULL, + .request_init = auth_glusterfs_v2_request_init, + .authenticate = auth_glusterfs_v2_authenticate}; + +rpcsvc_auth_t rpcsvc_auth_glusterfs_v2 = {.authname = "AUTH_GLUSTERFS-v2", + .authnum = AUTH_GLUSTERFS_v2, + .authops = &auth_glusterfs_ops_v2, + .authprivate = NULL}; + +rpcsvc_auth_t * +rpcsvc_auth_glusterfs_v2_init(rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_glusterfs_v2; +} + +/* V3 */ + +ssize_t +xdr_to_glusterfs_auth_v3(char *buf, struct auth_glusterfs_params_v3 *req) +{ + XDR xdr; + ssize_t ret = -1; + + if ((!buf) || (!req)) + return -1; + + xdrmem_create(&xdr, buf, GF_MAX_AUTH_BYTES, XDR_DECODE); + if (!xdr_auth_glusterfs_params_v3(&xdr, req)) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs v3 parameters"); + ret = -1; + goto ret; + } + + ret = (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)); +ret: + return ret; +} + +int +auth_glusterfs_v3_request_init(rpcsvc_request_t *req, void *priv) +{ + return 0; +} + +int +auth_glusterfs_v3_authenticate(rpcsvc_request_t *req, void *priv) +{ + struct auth_glusterfs_params_v3 au = { + 0, + }; + int ret = RPCSVC_AUTH_REJECT; + int i = 0; + int max_groups = 0; + int max_lk_owner_len = 0; + + if (!req) + return ret; + + ret = xdr_to_glusterfs_auth_v3(req->cred.authdata, &au); + if (ret == -1) { + gf_log("", GF_LOG_WARNING, "failed to decode glusterfs credentials"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + req->pid = au.pid; + req->uid = au.uid; + req->gid = au.gid; + req->lk_owner.len = au.lk_owner.lk_owner_len; + req->auxgidcount = au.groups.groups_len; + + /* the number of groups and size of lk_owner depend on each other */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS(req->lk_owner.len, + AUTH_GLUSTERFS_v3); + max_lk_owner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER(req->auxgidcount, + AUTH_GLUSTERFS_v3); + + if (req->auxgidcount > max_groups) { + gf_log("", GF_LOG_WARNING, + "more than max aux gids found (%d) , truncating it " + "to %d and continuing", + au.groups.groups_len, max_groups); + req->auxgidcount = max_groups; + } + + if (req->lk_owner.len > max_lk_owner_len) { + gf_log("", GF_LOG_WARNING, + "lkowner field to big (%d), depends on the number of " + "groups (%d), failing authentication", + req->lk_owner.len, req->auxgidcount); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + if (req->auxgidcount > SMALL_GROUP_COUNT) { + req->auxgidlarge = GF_CALLOC(req->auxgidcount, sizeof(req->auxgids[0]), + gf_common_mt_auxgids); + req->auxgids = req->auxgidlarge; + } else { + req->auxgids = req->auxgidsmall; + } + + if (!req->auxgids) { + gf_log("auth-glusterfs-v2", GF_LOG_WARNING, "cannot allocate gid list"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + for (i = 0; i < req->auxgidcount; ++i) + req->auxgids[i] = au.groups.groups_val[i]; + + for (i = 0; i < au.lk_owner.lk_owner_len; ++i) + req->lk_owner.data[i] = au.lk_owner.lk_owner_val[i]; + + /* All new things, starting glusterfs-4.0.0 */ + req->flags = au.flags; + req->ctime.tv_sec = au.ctime_sec; + req->ctime.tv_nsec = au.ctime_nsec; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "Auth Info: pid: %u, uid: %d" + ", gid: %d, owner: %s, flags: %d", + req->pid, req->uid, req->gid, lkowner_utoa(&req->lk_owner), + req->flags); + ret = RPCSVC_AUTH_ACCEPT; +err: + /* TODO: instead use alloca() for these variables */ + free(au.groups.groups_val); + free(au.lk_owner.lk_owner_val); + + return ret; +} + +rpcsvc_auth_ops_t auth_glusterfs_ops_v3 = { + .transport_init = NULL, + .request_init = auth_glusterfs_v3_request_init, + .authenticate = auth_glusterfs_v3_authenticate}; + +rpcsvc_auth_t rpcsvc_auth_glusterfs_v3 = {.authname = "AUTH_GLUSTERFS-v3", + .authnum = AUTH_GLUSTERFS_v3, + .authops = &auth_glusterfs_ops_v3, + .authprivate = NULL}; + +rpcsvc_auth_t * +rpcsvc_auth_glusterfs_v3_init(rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_glusterfs_v3; +} diff --git a/rpc/rpc-lib/src/auth-null.c b/rpc/rpc-lib/src/auth-null.c new file mode 100644 index 00000000000..6d059b9da50 --- /dev/null +++ b/rpc/rpc-lib/src/auth-null.c @@ -0,0 +1,40 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#include <glusterfs/dict.h> + +int +auth_null_request_init(rpcsvc_request_t *req, void *priv) +{ + return 0; +} + +int +auth_null_authenticate(rpcsvc_request_t *req, void *priv) +{ + /* Always succeed. */ + return RPCSVC_AUTH_ACCEPT; +} + +rpcsvc_auth_ops_t auth_null_ops = {.transport_init = NULL, + .request_init = auth_null_request_init, + .authenticate = auth_null_authenticate}; + +rpcsvc_auth_t rpcsvc_auth_null = {.authname = "AUTH_NULL", + .authnum = AUTH_NULL, + .authops = &auth_null_ops, + .authprivate = NULL}; + +rpcsvc_auth_t * +rpcsvc_auth_null_init(rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_null; +} diff --git a/rpc/rpc-lib/src/auth-unix.c b/rpc/rpc-lib/src/auth-unix.c new file mode 100644 index 00000000000..61d475a5e84 --- /dev/null +++ b/rpc/rpc-lib/src/auth-unix.c @@ -0,0 +1,66 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#include <glusterfs/dict.h> +#include "xdr-rpc.h" + +int +auth_unix_request_init(rpcsvc_request_t *req, void *priv) +{ + return 0; +} + +int +auth_unix_authenticate(rpcsvc_request_t *req, void *priv) +{ + int ret = RPCSVC_AUTH_REJECT; + struct authunix_parms aup; + char machname[MAX_MACHINE_NAME]; + + if (!req) + return ret; + + req->auxgids = req->auxgidsmall; + ret = xdr_to_auth_unix_cred(req->cred.authdata, req->cred.datalen, &aup, + machname, req->auxgids); + if (ret == -1) { + gf_log("", GF_LOG_WARNING, "failed to decode unix credentials"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + req->uid = aup.aup_uid; + req->gid = aup.aup_gid; + req->auxgidcount = aup.aup_len; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "Auth Info: machine name: %s, uid: %d" + ", gid: %d", + machname, req->uid, req->gid); + ret = RPCSVC_AUTH_ACCEPT; +err: + return ret; +} + +rpcsvc_auth_ops_t auth_unix_ops = {.transport_init = NULL, + .request_init = auth_unix_request_init, + .authenticate = auth_unix_authenticate}; + +rpcsvc_auth_t rpcsvc_auth_unix = {.authname = "AUTH_UNIX", + .authnum = AUTH_UNIX, + .authops = &auth_unix_ops, + .authprivate = NULL}; + +rpcsvc_auth_t * +rpcsvc_auth_unix_init(rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_unix; +} diff --git a/rpc/rpc-lib/src/autoscale-threads.c b/rpc/rpc-lib/src/autoscale-threads.c new file mode 100644 index 00000000000..a954ae7a27a --- /dev/null +++ b/rpc/rpc-lib/src/autoscale-threads.c @@ -0,0 +1,22 @@ +/* + Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include <glusterfs/gf-event.h> +#include "rpcsvc.h" + +void +rpcsvc_autoscale_threads(glusterfs_ctx_t *ctx, rpcsvc_t *rpc, int incr) +{ + struct event_pool *pool = ctx->event_pool; + int thread_count = pool->eventthreadcount; + + pool->auto_thread_count += incr; + (void)gf_event_reconfigure_threads(pool, thread_count + incr); +} diff --git a/rpc/rpc-lib/src/libgfrpc.sym b/rpc/rpc-lib/src/libgfrpc.sym new file mode 100644 index 00000000000..e026d80259b --- /dev/null +++ b/rpc/rpc-lib/src/libgfrpc.sym @@ -0,0 +1,68 @@ +is_rpc_clnt_disconnected +rpcclnt_cbk_program_register +rpc_clnt_cleanup_and_start +rpc_clnt_connection_cleanup +rpc_clnt_disable +rpc_clnt_new +rpc_clnt_reconfig +rpc_clnt_reconnect +rpc_clnt_reconnect_cleanup +rpc_clnt_ref +rpc_clnt_register_notify +rpc_clnt_start +rpc_clnt_submit +rpc_clnt_unref +rpc_reply_to_xdr +rpcsvc_auth_array +rpcsvc_auth_check +rpcsvc_auth_reconf +rpcsvc_auth_unix_auxgids +rpcsvc_callback_submit +rpcsvc_create_listener +rpcsvc_create_listeners +rpcsvc_drc_init +rpcsvc_drc_priv +rpcsvc_drc_reconfigure +rpcsvc_get_program_vector_sizer +rpcsvc_init +rpcsvc_destroy +rpcsvc_init_options +rpcsvc_listener_destroy +rpcsvc_program_register +rpcsvc_program_register_portmap +rpcsvc_program_register_rpcbind6 +rpcsvc_program_unregister +rpcsvc_program_unregister_portmap +rpcsvc_program_unregister_rpcbind6 +rpcsvc_reconfigure_options +rpcsvc_register_notify +rpcsvc_register_portmap_enabled +rpcsvc_request_submit +rpcsvc_set_outstanding_rpc_limit +rpcsvc_set_throttle_on +rpcsvc_submit_generic +rpcsvc_submit_message +rpcsvc_transport_peeraddr +rpcsvc_transport_peername +rpcsvc_transport_privport_check +rpcsvc_transport_unix_options_build +rpcsvc_transport_volume_allowed +rpcsvc_transport_connect +rpcsvc_transport_getpeeraddr +rpcsvc_unregister_notify +rpcsvc_volume_allowed +rpc_transport_count +rpc_transport_connect +rpc_transport_disconnect +rpc_transport_get_peeraddr +rpc_transport_inet_options_build +rpc_transport_keepalive_options_set +rpc_transport_notify +rpc_transport_pollin_alloc +rpc_transport_pollin_destroy +rpc_transport_ref +rpc_transport_unix_options_build +rpc_transport_unref +rpc_clnt_mgmt_pmap_signout +rpcsvc_autoscale_threads +rpcsvc_statedump diff --git a/rpc/rpc-lib/src/mgmt-pmap.c b/rpc/rpc-lib/src/mgmt-pmap.c new file mode 100644 index 00000000000..25a7148e5a3 --- /dev/null +++ b/rpc/rpc-lib/src/mgmt-pmap.c @@ -0,0 +1,147 @@ +/* + Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "portmap-xdr.h" +#include "protocol-common.h" +#include "rpc-clnt.h" +#include "xdr-generic.h" + +/* Defining a minimal RPC client program for portmap signout + */ +char *clnt_pmap_signout_procs[GF_PMAP_MAXVALUE] = { + [GF_PMAP_SIGNOUT] = "SIGNOUT", +}; + +rpc_clnt_prog_t clnt_pmap_signout_prog = { + .progname = "Gluster Portmap", + .prognum = GLUSTER_PMAP_PROGRAM, + .progver = GLUSTER_PMAP_VERSION, + .procnames = clnt_pmap_signout_procs, +}; + +static int +mgmt_pmap_signout_cbk(struct rpc_req *req, struct iovec *iov, int count, + void *myframe) +{ + pmap_signout_rsp rsp = { + 0, + }; + int ret = 0; + call_frame_t *frame = NULL; + + frame = myframe; + if (-1 == req->rpc_status) { + rsp.op_ret = -1; + rsp.op_errno = EINVAL; + goto out; + } + + ret = xdr_to_generic(*iov, &rsp, (xdrproc_t)xdr_pmap_signout_rsp); + if (ret < 0) { + gf_log(THIS->name, GF_LOG_ERROR, "XDR decoding failed"); + rsp.op_ret = -1; + rsp.op_errno = EINVAL; + goto out; + } + + if (-1 == rsp.op_ret) { + gf_log(THIS->name, GF_LOG_ERROR, + "failed to register the port with glusterd"); + goto out; + } +out: + if (frame) { + STACK_DESTROY(frame->root); + } + + return 0; +} + +int +rpc_clnt_mgmt_pmap_signout(glusterfs_ctx_t *ctx, char *brickname) +{ + int ret = 0; + pmap_signout_req req = { + 0, + }; + call_frame_t *frame = NULL; + cmd_args_t *cmd_args = NULL; + char brick_name[PATH_MAX] = { + 0, + }; + struct iovec iov = { + 0, + }; + struct iobuf *iobuf = NULL; + struct iobref *iobref = NULL; + ssize_t xdr_size = 0; + + frame = create_frame(THIS, ctx->pool); + cmd_args = &ctx->cmd_args; + + if (!cmd_args->brick_port && (!cmd_args->brick_name || !brickname)) { + gf_log("fsd-mgmt", GF_LOG_DEBUG, + "portmapper signout arguments not given"); + goto out; + } + + if (cmd_args->volfile_server_transport && + !strcmp(cmd_args->volfile_server_transport, "rdma")) { + snprintf(brick_name, sizeof(brick_name), "%s.rdma", + cmd_args->brick_name); + req.brick = brick_name; + } else { + if (brickname) + req.brick = brickname; + else + req.brick = cmd_args->brick_name; + } + + req.port = cmd_args->brick_port; + req.rdma_port = cmd_args->brick_port2; + + /* mgmt_submit_request is not available in libglusterfs. + * Need to serialize and submit manually. + */ + iobref = iobref_new(); + if (!iobref) { + goto out; + } + + xdr_size = xdr_sizeof((xdrproc_t)xdr_pmap_signout_req, &req); + iobuf = iobuf_get2(ctx->iobuf_pool, xdr_size); + if (!iobuf) { + goto out; + }; + + iobref_add(iobref, iobuf); + + iov.iov_base = iobuf->ptr; + iov.iov_len = iobuf_pagesize(iobuf); + + /* Create the xdr payload */ + ret = xdr_serialize_generic(iov, &req, (xdrproc_t)xdr_pmap_signout_req); + if (ret == -1) { + gf_log(THIS->name, GF_LOG_WARNING, "failed to create XDR payload"); + goto out; + } + iov.iov_len = ret; + + ret = rpc_clnt_submit(ctx->mgmt, &clnt_pmap_signout_prog, GF_PMAP_SIGNOUT, + mgmt_pmap_signout_cbk, &iov, 1, NULL, 0, iobref, + frame, NULL, 0, NULL, 0, NULL); +out: + if (iobref) + iobref_unref(iobref); + + if (iobuf) + iobuf_unref(iobuf); + return ret; +} diff --git a/rpc/rpc-lib/src/protocol-common.h b/rpc/rpc-lib/src/protocol-common.h new file mode 100644 index 00000000000..0cb5862e9a9 --- /dev/null +++ b/rpc/rpc-lib/src/protocol-common.h @@ -0,0 +1,382 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _PROTOCOL_COMMON_H +#define _PROTOCOL_COMMON_H + +enum gf_fop_procnum { + GFS3_OP_NULL, /* 0 */ + GFS3_OP_STAT, + GFS3_OP_READLINK, + GFS3_OP_MKNOD, + GFS3_OP_MKDIR, + GFS3_OP_UNLINK, + GFS3_OP_RMDIR, + GFS3_OP_SYMLINK, + GFS3_OP_RENAME, + GFS3_OP_LINK, + GFS3_OP_TRUNCATE, + GFS3_OP_OPEN, + GFS3_OP_READ, + GFS3_OP_WRITE, + GFS3_OP_STATFS, + GFS3_OP_FLUSH, + GFS3_OP_FSYNC, + GFS3_OP_SETXATTR, + GFS3_OP_GETXATTR, + GFS3_OP_REMOVEXATTR, + GFS3_OP_OPENDIR, + GFS3_OP_FSYNCDIR, + GFS3_OP_ACCESS, + GFS3_OP_CREATE, + GFS3_OP_FTRUNCATE, + GFS3_OP_FSTAT, + GFS3_OP_LK, + GFS3_OP_LOOKUP, + GFS3_OP_READDIR, + GFS3_OP_INODELK, + GFS3_OP_FINODELK, + GFS3_OP_ENTRYLK, + GFS3_OP_FENTRYLK, + GFS3_OP_XATTROP, + GFS3_OP_FXATTROP, + GFS3_OP_FGETXATTR, + GFS3_OP_FSETXATTR, + GFS3_OP_RCHECKSUM, + GFS3_OP_SETATTR, + GFS3_OP_FSETATTR, + GFS3_OP_READDIRP, + GFS3_OP_RELEASE, + GFS3_OP_RELEASEDIR, + GFS3_OP_FREMOVEXATTR, + GFS3_OP_FALLOCATE, + GFS3_OP_DISCARD, + GFS3_OP_ZEROFILL, + GFS3_OP_IPC, + GFS3_OP_SEEK, + GFS3_OP_COMPOUND, + GFS3_OP_LEASE, + GFS3_OP_GETACTIVELK, + GFS3_OP_SETACTIVELK, + GFS3_OP_ICREATE, + GFS3_OP_NAMELINK, + GFS3_OP_PUT, + GFS3_OP_COPY_FILE_RANGE, + GFS3_OP_MAXVALUE, +}; + +enum gf_handshake_procnum { + GF_HNDSK_NULL, + GF_HNDSK_SETVOLUME, + GF_HNDSK_GETSPEC, + GF_HNDSK_PING, + GF_HNDSK_SET_LK_VER, + GF_HNDSK_EVENT_NOTIFY, + GF_HNDSK_GET_VOLUME_INFO, + GF_HNDSK_GET_SNAPSHOT_INFO, + GF_HNDSK_MAXVALUE, +}; + +enum gf_pmap_procnum { + GF_PMAP_NULL = 0, + GF_PMAP_PORTBYBRICK, + GF_PMAP_BRICKBYPORT, + /* + * SIGNUP is not used, and shouldn't be used. It was kept here only + * to avoid changing the numbers for things that come after it in this + * list. + */ + GF_PMAP_SIGNUP, + GF_PMAP_SIGNIN, + GF_PMAP_SIGNOUT, + GF_PMAP_MAXVALUE, +}; + +enum gf_aggregator_procnum { + GF_AGGREGATOR_NULL = 0, + GF_AGGREGATOR_LOOKUP, + GF_AGGREGATOR_GETLIMIT, + GF_AGGREGATOR_MAXVALUE, +}; + +enum gf_pmap_port_type { + GF_PMAP_PORT_FREE = 0, + GF_PMAP_PORT_FOREIGN, /* it actually means, not sure who is using it, but it + is in-use */ + GF_PMAP_PORT_LEASED, + GF_PMAP_PORT_ANY, + GF_PMAP_PORT_BRICKSERVER, /* port used by brick process */ +}; +typedef enum gf_pmap_port_type gf_pmap_port_type_t; + +enum gf_probe_resp { + GF_PROBE_SUCCESS, + GF_PROBE_LOCALHOST, + GF_PROBE_FRIEND, + GF_PROBE_ANOTHER_CLUSTER, + GF_PROBE_VOLUME_CONFLICT, + GF_PROBE_SAME_UUID, + GF_PROBE_UNKNOWN_PEER, + GF_PROBE_ADD_FAILED, + GF_PROBE_QUORUM_NOT_MET, + GF_PROBE_MISSED_SNAP_CONFLICT, + GF_PROBE_SNAP_CONFLICT, + GF_PROBE_FRIEND_DETACHING, +}; + +enum gf_deprobe_resp { + GF_DEPROBE_SUCCESS, + GF_DEPROBE_LOCALHOST, + GF_DEPROBE_NOT_FRIEND, + GF_DEPROBE_BRICK_EXIST, + GF_DEPROBE_FRIEND_DOWN, + GF_DEPROBE_QUORUM_NOT_MET, + GF_DEPROBE_FRIEND_DETACHING, + GF_DEPROBE_SNAP_BRICK_EXIST, +}; + +enum gf_cbk_procnum { + GF_CBK_NULL = 0, + GF_CBK_FETCHSPEC, + GF_CBK_INO_FLUSH, + GF_CBK_EVENT_NOTIFY, + GF_CBK_GET_SNAPS, + GF_CBK_CACHE_INVALIDATION, + GF_CBK_CHILD_UP, + GF_CBK_CHILD_DOWN, + GF_CBK_RECALL_LEASE, + GF_CBK_STATEDUMP, + GF_CBK_INODELK_CONTENTION, + GF_CBK_ENTRYLK_CONTENTION, + GF_CBK_MAXVALUE, +}; + +enum gluster_cli_procnum { + GLUSTER_CLI_NULL, /* 0 */ + GLUSTER_CLI_PROBE, + GLUSTER_CLI_DEPROBE, + GLUSTER_CLI_LIST_FRIENDS, + GLUSTER_CLI_CREATE_VOLUME, + GLUSTER_CLI_GET_VOLUME, + GLUSTER_CLI_GET_NEXT_VOLUME, + GLUSTER_CLI_DELETE_VOLUME, + GLUSTER_CLI_START_VOLUME, + GLUSTER_CLI_STOP_VOLUME, + GLUSTER_CLI_RENAME_VOLUME, + GLUSTER_CLI_DEFRAG_VOLUME, + GLUSTER_CLI_SET_VOLUME, + GLUSTER_CLI_ADD_BRICK, + GLUSTER_CLI_REMOVE_BRICK, + GLUSTER_CLI_REPLACE_BRICK, + GLUSTER_CLI_LOG_ROTATE, + GLUSTER_CLI_GETSPEC, + GLUSTER_CLI_PMAP_PORTBYBRICK, + GLUSTER_CLI_SYNC_VOLUME, + GLUSTER_CLI_RESET_VOLUME, + GLUSTER_CLI_FSM_LOG, + GLUSTER_CLI_GSYNC_SET, + GLUSTER_CLI_PROFILE_VOLUME, + GLUSTER_CLI_QUOTA, + GLUSTER_CLI_TOP_VOLUME, + GLUSTER_CLI_GETWD, + GLUSTER_CLI_STATUS_VOLUME, + GLUSTER_CLI_STATUS_ALL, + GLUSTER_CLI_MOUNT, + GLUSTER_CLI_UMOUNT, + GLUSTER_CLI_HEAL_VOLUME, + GLUSTER_CLI_STATEDUMP_VOLUME, + GLUSTER_CLI_LIST_VOLUME, + GLUSTER_CLI_CLRLOCKS_VOLUME, + GLUSTER_CLI_UUID_RESET, + GLUSTER_CLI_UUID_GET, + GLUSTER_CLI_COPY_FILE, + GLUSTER_CLI_SYS_EXEC, + GLUSTER_CLI_SNAP, + GLUSTER_CLI_BARRIER_VOLUME, + GLUSTER_CLI_GET_VOL_OPT, + GLUSTER_CLI_GANESHA, + GLUSTER_CLI_BITROT, + GLUSTER_CLI_ATTACH_TIER, + GLUSTER_CLI_TIER, + GLUSTER_CLI_GET_STATE, + GLUSTER_CLI_RESET_BRICK, + GLUSTER_CLI_REMOVE_TIER_BRICK, + GLUSTER_CLI_ADD_TIER_BRICK, + GLUSTER_CLI_MAXVALUE, +}; + +enum glusterd_mgmt_procnum { + GLUSTERD_MGMT_NULL, /* 0 */ + GLUSTERD_MGMT_CLUSTER_LOCK, + GLUSTERD_MGMT_CLUSTER_UNLOCK, + GLUSTERD_MGMT_STAGE_OP, + GLUSTERD_MGMT_COMMIT_OP, + GLUSTERD_MGMT_MAXVALUE, +}; + +enum glusterd_friend_procnum { + GLUSTERD_FRIEND_NULL, /* 0 */ + GLUSTERD_PROBE_QUERY, + GLUSTERD_FRIEND_ADD, + GLUSTERD_FRIEND_REMOVE, + GLUSTERD_FRIEND_UPDATE, + GLUSTERD_FRIEND_MAXVALUE, +}; + +enum glusterd_brick_procnum { + GLUSTERD_BRICK_NULL, /* 0 */ + GLUSTERD_BRICK_TERMINATE, + GLUSTERD_BRICK_XLATOR_INFO, + GLUSTERD_BRICK_XLATOR_OP, + GLUSTERD_BRICK_STATUS, + GLUSTERD_BRICK_OP, + GLUSTERD_BRICK_XLATOR_DEFRAG, + GLUSTERD_NODE_PROFILE, + GLUSTERD_NODE_STATUS, + GLUSTERD_VOLUME_BARRIER_OP, + GLUSTERD_BRICK_BARRIER, + GLUSTERD_NODE_BITROT, + GLUSTERD_BRICK_ATTACH, + GLUSTERD_DUMP_METRICS, + GLUSTERD_SVC_ATTACH, + GLUSTERD_SVC_DETACH, + GLUSTERD_BRICK_MAXVALUE, +}; + +enum glusterd_mgmt_hndsk_procnum { + GD_MGMT_HNDSK_NULL, + GD_MGMT_HNDSK_VERSIONS, + GD_MGMT_HNDSK_VERSIONS_ACK, + GD_MGMT_HNDSK_MAXVALUE, +}; + +typedef enum { + GF_SHD_OP_INVALID, + GF_SHD_OP_HEAL_INDEX, + GF_SHD_OP_HEAL_FULL, + GF_SHD_OP_INDEX_SUMMARY, + GF_SHD_OP_HEALED_FILES, + GF_SHD_OP_HEAL_FAILED_FILES, + GF_SHD_OP_SPLIT_BRAIN_FILES, + GF_SHD_OP_STATISTICS, + GF_SHD_OP_STATISTICS_HEAL_COUNT, + GF_SHD_OP_STATISTICS_HEAL_COUNT_PER_REPLICA, + GF_SHD_OP_SBRAIN_HEAL_FROM_BIGGER_FILE, + GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK, + GF_SHD_OP_HEAL_ENABLE, + GF_SHD_OP_HEAL_DISABLE, + GF_SHD_OP_SBRAIN_HEAL_FROM_LATEST_MTIME, + GF_SHD_OP_GRANULAR_ENTRY_HEAL_ENABLE, + GF_SHD_OP_GRANULAR_ENTRY_HEAL_DISABLE, + GF_SHD_OP_HEAL_SUMMARY, +} gf_xl_afr_op_t; + +struct gf_gsync_detailed_status_ { + char node[NAME_MAX]; + char master[NAME_MAX]; + char brick[PATH_MAX]; + char slave_user[NAME_MAX]; + char slave[NAME_MAX]; + char slave_node[NAME_MAX]; + char worker_status[NAME_MAX]; + char crawl_status[NAME_MAX]; + char last_synced[NAME_MAX]; + char last_synced_utc[NAME_MAX]; + char entry[NAME_MAX]; + char data[NAME_MAX]; + char meta[NAME_MAX]; + char failures[NAME_MAX]; + char checkpoint_time[NAME_MAX]; + char checkpoint_time_utc[NAME_MAX]; + char checkpoint_completed[NAME_MAX]; + char checkpoint_completion_time[NAME_MAX]; + char checkpoint_completion_time_utc[NAME_MAX]; + char brick_host_uuid[NAME_MAX]; + char slavekey[NAME_MAX]; + char session_slave[NAME_MAX]; +}; + +enum glusterd_mgmt_v3_procnum { + GLUSTERD_MGMT_V3_NULL, /* 0 */ + GLUSTERD_MGMT_V3_LOCK, + GLUSTERD_MGMT_V3_PRE_VALIDATE, + GLUSTERD_MGMT_V3_BRICK_OP, + GLUSTERD_MGMT_V3_COMMIT, + GLUSTERD_MGMT_V3_POST_COMMIT, + GLUSTERD_MGMT_V3_POST_VALIDATE, + GLUSTERD_MGMT_V3_UNLOCK, + GLUSTERD_MGMT_V3_MAXVALUE, +}; + +typedef struct gf_gsync_detailed_status_ gf_gsync_status_t; + +enum gf_get_volume_info_type { + GF_GET_VOLUME_NONE, /* 0 */ + GF_GET_VOLUME_UUID +}; + +typedef enum gf_get_volume_info_type gf_get_volume_info_type; + +enum gf_get_snapshot_info_type { + GF_GET_SNAPSHOT_LIST, +}; +typedef enum gf_get_snapshot_info_type gf_get_snapshot_info_type; + +enum gf_getspec_flags_type { GF_GETSPEC_FLAG_SERVERS_LIST = 1 }; +typedef enum gf_getspec_flags_type gf_getspec_flags_type; + +#define GLUSTER_HNDSK_PROGRAM 14398633 /* Completely random */ +#define GLUSTER_HNDSK_VERSION 2 /* 0.0.2 */ + +#define GLUSTER_PMAP_PROGRAM 34123456 +#define GLUSTER_PMAP_VERSION 1 + +#define GLUSTER_CBK_PROGRAM 52743234 /* Completely random */ +#define GLUSTER_CBK_VERSION 1 /* 0.0.1 */ + +#define GLUSTER_FOP_PROGRAM 1298437 /* Completely random */ +#define GLUSTER_FOP_VERSION 330 /* 3.3.0 */ +#define GLUSTER_FOP_PROCCNT GFS3_OP_MAXVALUE + +#define GLUSTER_FOP_VERSION_v2 400 /* 4.0.0 */ + +/* Aggregator */ +#define GLUSTER_AGGREGATOR_PROGRAM 29852134 /* Completely random */ +#define GLUSTER_AGGREGATOR_VERSION 1 + +/* Second version */ +#define GD_MGMT_PROGRAM 1238433 /* Completely random */ +#define GD_MGMT_VERSION 2 /* 0.0.2 */ + +#define GD_FRIEND_PROGRAM 1238437 /* Completely random */ +#define GD_FRIEND_VERSION 2 /* 0.0.2 */ + +#define GLUSTER_CLI_PROGRAM 1238463 /* Completely random */ +#define GLUSTER_CLI_VERSION 2 /* 0.0.2 */ + +#define GD_BRICK_PROGRAM 4867634 /*Completely random*/ +#define GD_BRICK_VERSION 2 + +/* Third version */ +#define GD_MGMT_V3_VERSION 3 + +/* OP-VERSION handshake */ +#define GD_MGMT_HNDSK_PROGRAM 1239873 /* Completely random */ +#define GD_MGMT_HNDSK_VERSION 1 + +#define GD_VOLUME_NAME_MAX \ + ((NAME_MAX + 1) - 5) /* Maximum size of volume name */ +#define GD_VOLUME_NAME_MAX_TIER \ + (GD_VOLUME_NAME_MAX + 5) /* +5 needed for '-hot' \ + and '-cold' suffixes*/ + +#define GLUSTER_PROCESS_UUID_FMT \ + "CTX_ID:%s-GRAPH_ID:%d-PID:%d-HOST:%s-PC_NAME:%s-RECON_NO:%s" +#endif /* !_PROTOCOL_COMMON_H */ diff --git a/rpc/rpc-lib/src/rpc-clnt-ping.c b/rpc/rpc-lib/src/rpc-clnt-ping.c new file mode 100644 index 00000000000..31f17841bea --- /dev/null +++ b/rpc/rpc-lib/src/rpc-clnt-ping.c @@ -0,0 +1,357 @@ +/* + Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpc-clnt.h" +#include "rpc-clnt-ping.h" +#include <glusterfs/byte-order.h> +#include "xdr-rpcclnt.h" +#include "rpc-transport.h" +#include "protocol-common.h" +#include <glusterfs/mem-pool.h> +#include "xdr-rpc.h" +#include "rpc-common-xdr.h" +#include <glusterfs/timespec.h> + +char *clnt_ping_procs[GF_DUMP_MAXVALUE] = { + [GF_DUMP_PING] = "NULL", +}; +struct rpc_clnt_program clnt_ping_prog = { + .progname = "GF-DUMP", + .prognum = GLUSTER_DUMP_PROGRAM, + .progver = GLUSTER_DUMP_VERSION, + .procnames = clnt_ping_procs, +}; + +struct ping_local { + struct rpc_clnt *rpc; + struct timespec submit_time; +}; + +/* Must be called under conn->lock */ +static int +__rpc_clnt_rearm_ping_timer(struct rpc_clnt *rpc, gf_timer_cbk_t cbk) +{ + rpc_clnt_connection_t *conn = &rpc->conn; + rpc_transport_t *trans = conn->trans; + struct timespec timeout = { + 0, + }; + gf_timer_t *timer = NULL; + + if (conn->ping_timer) { + gf_log_callingfn("", GF_LOG_CRITICAL, + "%s: ping timer event already scheduled", + conn->trans->peerinfo.identifier); + return -1; + } + + timeout.tv_sec = conn->ping_timeout; + timeout.tv_nsec = 0; + + rpc_clnt_ref(rpc); + timer = gf_timer_call_after(rpc->ctx, timeout, cbk, (void *)rpc); + if (timer == NULL) { + gf_log(trans->name, GF_LOG_WARNING, "unable to setup ping timer"); + + /* This unref can't be the last. We just took a ref few lines + * above. So this can be performed under conn->lock. */ + rpc_clnt_unref(rpc); + conn->ping_started = 0; + return -1; + } + + conn->ping_timer = timer; + conn->ping_started = 1; + return 0; +} + +/* Must be called under conn->lock */ +int +rpc_clnt_remove_ping_timer_locked(struct rpc_clnt *rpc) +{ + rpc_clnt_connection_t *conn = &rpc->conn; + gf_timer_t *timer = NULL; + + if (conn->ping_timer) { + timer = conn->ping_timer; + conn->ping_timer = NULL; + gf_timer_call_cancel(rpc->ctx, timer); + conn->ping_started = 0; + return 1; + } + + /* This is to account for rpc_clnt_disable that might have set + * conn->trans to NULL. */ + if (conn->trans) + gf_log_callingfn("", GF_LOG_DEBUG, + "%s: ping timer event " + "already removed", + conn->trans->peerinfo.identifier); + + return 0; +} + +static void +rpc_clnt_start_ping(void *rpc_ptr); + +void +rpc_clnt_ping_timer_expired(void *rpc_ptr) +{ + struct rpc_clnt *rpc = NULL; + rpc_transport_t *trans = NULL; + rpc_clnt_connection_t *conn = NULL; + int disconnect = 0; + struct timespec current = { + 0, + }; + int unref = 0; + + rpc = (struct rpc_clnt *)rpc_ptr; + conn = &rpc->conn; + trans = conn->trans; + + if (!trans) { + gf_log("ping-timer", GF_LOG_WARNING, "transport not initialized"); + goto out; + } + + timespec_now_realtime(¤t); + pthread_mutex_lock(&conn->lock); + { + unref = rpc_clnt_remove_ping_timer_locked(rpc); + + if (((current.tv_sec - conn->last_received.tv_sec) < + conn->ping_timeout) || + ((current.tv_sec - conn->last_sent.tv_sec) < conn->ping_timeout)) { + gf_log(trans->name, GF_LOG_TRACE, + "ping timer expired but transport activity " + "detected - not bailing transport"); + if (__rpc_clnt_rearm_ping_timer(rpc, rpc_clnt_ping_timer_expired) == + -1) { + gf_log(trans->name, GF_LOG_WARNING, + "unable to setup ping timer"); + } + } else { + conn->ping_started = 0; + disconnect = 1; + } + } + pthread_mutex_unlock(&conn->lock); + + if (unref) + rpc_clnt_unref(rpc); + + if (disconnect) { + gf_log(trans->name, GF_LOG_CRITICAL, + "server %s has not responded in the last %d " + "seconds, disconnecting.", + trans->peerinfo.identifier, conn->ping_timeout); + + rpc_transport_disconnect(conn->trans, _gf_false); + } + +out: + return; +} + +int +rpc_clnt_ping_cbk(struct rpc_req *req, struct iovec *iov, int count, + void *myframe) +{ + struct ping_local *local = NULL; + xlator_t *this = NULL; + rpc_clnt_connection_t *conn = NULL; + call_frame_t *frame = NULL; + int unref = 0; + gf_boolean_t call_notify = _gf_false; + + struct timespec now; + struct timespec delta; + int64_t latency_msec = 0; + int ret = 0; + + if (!myframe) { + gf_log(THIS->name, GF_LOG_WARNING, "frame with the request is NULL"); + goto out; + } + + frame = myframe; + this = frame->this; + local = frame->local; + conn = &local->rpc->conn; + + timespec_now(&now); + timespec_sub(&local->submit_time, &now, &delta); + latency_msec = delta.tv_sec * 1000 + delta.tv_nsec / 1000000; + + gf_log(THIS->name, GF_LOG_DEBUG, "Ping latency is %" PRIu64 "ms", + latency_msec); + call_notify = _gf_true; + + pthread_mutex_lock(&conn->lock); + { + unref = rpc_clnt_remove_ping_timer_locked(local->rpc); + if (req->rpc_status == -1) { + conn->ping_started = 0; + pthread_mutex_unlock(&conn->lock); + if (unref) { + gf_log(this->name, GF_LOG_WARNING, + "socket or ib related error"); + + } else { + /* timer expired and transport bailed out */ + gf_log(this->name, GF_LOG_WARNING, "socket disconnected"); + } + goto after_unlock; + } + + if (__rpc_clnt_rearm_ping_timer(local->rpc, rpc_clnt_start_ping) == + -1) { + /* unlock before logging error */ + pthread_mutex_unlock(&conn->lock); + gf_log(this->name, GF_LOG_WARNING, "failed to set the ping timer"); + } else { + /* just unlock the mutex */ + pthread_mutex_unlock(&conn->lock); + } + } +after_unlock: + if (call_notify) { + ret = local->rpc->notifyfn(local->rpc, this, RPC_CLNT_PING, + (void *)(uintptr_t)latency_msec); + if (ret) { + gf_log(this->name, GF_LOG_WARNING, "RPC_CLNT_PING notify failed"); + } + } +out: + if (unref) + rpc_clnt_unref(local->rpc); + + if (frame) { + GF_FREE(frame->local); + frame->local = NULL; + STACK_DESTROY(frame->root); + } + return 0; +} + +int +rpc_clnt_ping(struct rpc_clnt *rpc) +{ + call_frame_t *frame = NULL; + int32_t ret = -1; + rpc_clnt_connection_t *conn = NULL; + struct ping_local *local = NULL; + + conn = &rpc->conn; + local = GF_MALLOC(sizeof(struct ping_local), gf_common_ping_local_t); + if (!local) + return ret; + frame = create_frame(THIS, THIS->ctx->pool); + if (!frame) { + GF_FREE(local); + return ret; + } + + local->rpc = rpc; + timespec_now(&local->submit_time); + frame->local = local; + + ret = rpc_clnt_submit(rpc, &clnt_ping_prog, GF_DUMP_PING, rpc_clnt_ping_cbk, + NULL, 0, NULL, 0, NULL, frame, NULL, 0, NULL, 0, + NULL); + if (ret) { + /* FIXME: should we free the frame here? Methinks so! */ + gf_log(THIS->name, GF_LOG_ERROR, "failed to start ping timer"); + } else { + /* ping successfully queued in list of saved frames + * for the connection*/ + pthread_mutex_lock(&conn->lock); + conn->pingcnt++; + pthread_mutex_unlock(&conn->lock); + } + + return ret; +} + +static void +rpc_clnt_start_ping(void *rpc_ptr) +{ + struct rpc_clnt *rpc = NULL; + rpc_clnt_connection_t *conn = NULL; + int frame_count = 0; + int unref = 0; + + rpc = (struct rpc_clnt *)rpc_ptr; + conn = &rpc->conn; + + if (conn->ping_timeout == 0) { + gf_log(THIS->name, GF_LOG_DEBUG, + "ping timeout is 0," + " returning"); + return; + } + + pthread_mutex_lock(&conn->lock); + { + unref = rpc_clnt_remove_ping_timer_locked(rpc); + + if (conn->saved_frames) { + GF_ASSERT(conn->saved_frames->count >= 0); + /* treat the case where conn->saved_frames is NULL + as no pending frames */ + frame_count = conn->saved_frames->count; + } + + if ((frame_count == 0) || !conn->connected) { + gf_log(THIS->name, GF_LOG_DEBUG, + "returning as transport is already disconnected" + " OR there are no frames (%d || %d)", + !conn->connected, frame_count); + + pthread_mutex_unlock(&conn->lock); + if (unref) + rpc_clnt_unref(rpc); + return; + } + + if (__rpc_clnt_rearm_ping_timer(rpc, rpc_clnt_ping_timer_expired) == + -1) { + gf_log(THIS->name, GF_LOG_WARNING, "unable to setup ping timer"); + pthread_mutex_unlock(&conn->lock); + if (unref) + rpc_clnt_unref(rpc); + return; + } + } + pthread_mutex_unlock(&conn->lock); + if (unref) + rpc_clnt_unref(rpc); + + rpc_clnt_ping(rpc); +} + +void +rpc_clnt_check_and_start_ping(struct rpc_clnt *rpc) +{ + char start_ping = 0; + + pthread_mutex_lock(&rpc->conn.lock); + { + if (!rpc->conn.ping_started) + start_ping = 1; + } + pthread_mutex_unlock(&rpc->conn.lock); + + if (start_ping) + rpc_clnt_start_ping((void *)rpc); + + return; +} diff --git a/rpc/rpc-lib/src/rpc-clnt-ping.h b/rpc/rpc-lib/src/rpc-clnt-ping.h new file mode 100644 index 00000000000..e5466a828c2 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-clnt-ping.h @@ -0,0 +1,16 @@ +/* + Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +struct rpc_clnt; +#define RPC_DEFAULT_PING_TIMEOUT 30 +void +rpc_clnt_check_and_start_ping(struct rpc_clnt *rpc_ptr); +int +rpc_clnt_remove_ping_timer_locked(struct rpc_clnt *rpc); diff --git a/rpc/rpc-lib/src/rpc-clnt.c b/rpc/rpc-lib/src/rpc-clnt.c new file mode 100644 index 00000000000..517037c4a5d --- /dev/null +++ b/rpc/rpc-lib/src/rpc-clnt.c @@ -0,0 +1,1979 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#define RPC_CLNT_DEFAULT_REQUEST_COUNT 512 + +#include "rpc-clnt.h" +#include "rpc-clnt-ping.h" +#include <glusterfs/byte-order.h> +#include "xdr-rpcclnt.h" +#include "rpc-transport.h" +#include "protocol-common.h" +#include <glusterfs/mem-pool.h> +#include "xdr-rpc.h" +#include "rpc-common-xdr.h" + +void +rpc_clnt_reply_deinit(struct rpc_req *req, struct mem_pool *pool); + +struct saved_frame * +__saved_frames_get_timedout(struct saved_frames *frames, uint32_t timeout, + struct timeval *current) +{ + struct saved_frame *bailout_frame = NULL, *tmp = NULL; + + if (!list_empty(&frames->sf.list)) { + tmp = list_entry(frames->sf.list.next, typeof(*tmp), list); + if ((tmp->saved_at.tv_sec + timeout) <= current->tv_sec) { + bailout_frame = tmp; + list_del_init(&bailout_frame->list); + frames->count--; + } + } + + return bailout_frame; +} + +static int +_is_lock_fop(struct saved_frame *sframe) +{ + int fop = 0; + + if (SFRAME_GET_PROGNUM(sframe) == GLUSTER_FOP_PROGRAM && + SFRAME_GET_PROGVER(sframe) == GLUSTER_FOP_VERSION) + fop = SFRAME_GET_PROCNUM(sframe); + + return ((fop == GFS3_OP_LK) || (fop == GFS3_OP_INODELK) || + (fop == GFS3_OP_FINODELK) || (fop == GFS3_OP_ENTRYLK) || + (fop == GFS3_OP_FENTRYLK)); +} + +static struct saved_frame * +__saved_frames_put(struct saved_frames *frames, void *frame, + struct rpc_req *rpcreq) +{ + struct saved_frame *saved_frame = mem_get( + rpcreq->conn->rpc_clnt->saved_frames_pool); + + if (!saved_frame) { + goto out; + } + /* THIS should be saved and set back */ + + INIT_LIST_HEAD(&saved_frame->list); + + saved_frame->capital_this = THIS; + saved_frame->frame = frame; + saved_frame->rpcreq = rpcreq; + gettimeofday(&saved_frame->saved_at, NULL); + memset(&saved_frame->rsp, 0, sizeof(rpc_transport_rsp_t)); + + if (_is_lock_fop(saved_frame)) + list_add_tail(&saved_frame->list, &frames->lk_sf.list); + else + list_add_tail(&saved_frame->list, &frames->sf.list); + + frames->count++; + +out: + return saved_frame; +} + +static void +call_bail(void *data) +{ + rpc_transport_t *trans = NULL; + struct rpc_clnt *clnt = NULL; + rpc_clnt_connection_t *conn = NULL; + struct timeval current; + struct list_head list; + struct saved_frame *saved_frame = NULL; + struct saved_frame *trav = NULL; + struct saved_frame *tmp = NULL; + char frame_sent[GF_TIMESTR_SIZE] = { + 0, + }; + struct timespec timeout = { + 0, + }; + char peerid[UNIX_PATH_MAX] = {0}; + gf_boolean_t need_unref = _gf_false; + + GF_VALIDATE_OR_GOTO("client", data, out); + + clnt = data; + + conn = &clnt->conn; + pthread_mutex_lock(&conn->lock); + { + trans = conn->trans; + if (trans) { + (void)snprintf(peerid, sizeof(peerid), "%s", + conn->trans->peerinfo.identifier); + } + } + pthread_mutex_unlock(&conn->lock); + /*rpc_clnt_connection_cleanup will be unwinding all saved frames, + * bailed or otherwise*/ + if (!trans) + goto out; + + gettimeofday(¤t, NULL); + INIT_LIST_HEAD(&list); + + pthread_mutex_lock(&conn->lock); + { + /* Chaining to get call-always functionality from + call-once timer */ + if (conn->timer) { + timeout.tv_sec = 10; + timeout.tv_nsec = 0; + + /* Ref rpc as it's added to timer event queue */ + rpc_clnt_ref(clnt); + gf_timer_call_cancel(clnt->ctx, conn->timer); + conn->timer = gf_timer_call_after(clnt->ctx, timeout, call_bail, + (void *)clnt); + + if (conn->timer == NULL) { + gf_log(conn->name, GF_LOG_WARNING, + "Cannot create bailout timer for %s", peerid); + need_unref = _gf_true; + } + } + + do { + saved_frame = __saved_frames_get_timedout( + conn->saved_frames, conn->frame_timeout, ¤t); + if (saved_frame) + list_add(&saved_frame->list, &list); + + } while (saved_frame); + } + pthread_mutex_unlock(&conn->lock); + + if (list_empty(&list)) + goto out; + + list_for_each_entry_safe(trav, tmp, &list, list) + { + gf_time_fmt_tv(frame_sent, sizeof frame_sent, &trav->saved_at, + gf_timefmt_FT); + + gf_log(conn->name, GF_LOG_ERROR, + "bailing out frame type(%s), op(%s(%d)), xid = 0x%x, " + "unique = %" PRIu64 ", sent = %s, timeout = %d for %s", + trav->rpcreq->prog->progname, + (trav->rpcreq->prog->procnames) + ? trav->rpcreq->prog->procnames[trav->rpcreq->procnum] + : "--", + trav->rpcreq->procnum, trav->rpcreq->xid, + ((call_frame_t *)(trav->frame))->root->unique, frame_sent, + conn->frame_timeout, peerid); + + clnt = rpc_clnt_ref(clnt); + trav->rpcreq->rpc_status = -1; + trav->rpcreq->cbkfn(trav->rpcreq, NULL, 0, trav->frame); + + rpc_clnt_reply_deinit(trav->rpcreq, clnt->reqpool); + clnt = rpc_clnt_unref(clnt); + list_del_init(&trav->list); + mem_put(trav); + } +out: + rpc_clnt_unref(clnt); + if (need_unref) + rpc_clnt_unref(clnt); + return; +} + +/* to be called with conn->lock held */ +static struct saved_frame * +__save_frame(struct rpc_clnt *rpc_clnt, call_frame_t *frame, + struct rpc_req *rpcreq) +{ + rpc_clnt_connection_t *conn = &rpc_clnt->conn; + struct timespec timeout = { + 0, + }; + struct saved_frame *saved_frame = __saved_frames_put(conn->saved_frames, + frame, rpcreq); + + if (saved_frame == NULL) { + goto out; + } + + /* TODO: make timeout configurable */ + if (conn->timer == NULL) { + timeout.tv_sec = 10; + timeout.tv_nsec = 0; + rpc_clnt_ref(rpc_clnt); + conn->timer = gf_timer_call_after(rpc_clnt->ctx, timeout, call_bail, + (void *)rpc_clnt); + } + +out: + return saved_frame; +} + +struct saved_frames * +saved_frames_new(void) +{ + struct saved_frames *saved_frames = NULL; + + saved_frames = GF_CALLOC(1, sizeof(*saved_frames), + gf_common_mt_rpcclnt_savedframe_t); + if (!saved_frames) { + return NULL; + } + + INIT_LIST_HEAD(&saved_frames->sf.list); + INIT_LIST_HEAD(&saved_frames->lk_sf.list); + + return saved_frames; +} + +int +__saved_frame_copy(struct saved_frames *frames, int64_t callid, + struct saved_frame *saved_frame) +{ + struct saved_frame *tmp = NULL; + int ret = -1; + + if (!saved_frame) { + ret = 0; + goto out; + } + + list_for_each_entry(tmp, &frames->sf.list, list) + { + if (tmp->rpcreq->xid == callid) { + *saved_frame = *tmp; + ret = 0; + goto out; + } + } + + list_for_each_entry(tmp, &frames->lk_sf.list, list) + { + if (tmp->rpcreq->xid == callid) { + *saved_frame = *tmp; + ret = 0; + goto out; + } + } + +out: + return ret; +} + +struct saved_frame * +__saved_frame_get(struct saved_frames *frames, int64_t callid) +{ + struct saved_frame *saved_frame = NULL; + struct saved_frame *tmp = NULL; + + list_for_each_entry(tmp, &frames->sf.list, list) + { + if (tmp->rpcreq->xid == callid) { + list_del_init(&tmp->list); + frames->count--; + saved_frame = tmp; + goto out; + } + } + + list_for_each_entry(tmp, &frames->lk_sf.list, list) + { + if (tmp->rpcreq->xid == callid) { + list_del_init(&tmp->list); + frames->count--; + saved_frame = tmp; + goto out; + } + } + +out: + if (saved_frame) { + THIS = saved_frame->capital_this; + } + + return saved_frame; +} + +void +saved_frames_unwind(struct saved_frames *saved_frames) +{ + struct saved_frame *trav = NULL; + struct saved_frame *tmp = NULL; + char timestr[GF_TIMESTR_SIZE] = { + 0, + }; + + list_splice_init(&saved_frames->lk_sf.list, &saved_frames->sf.list); + + list_for_each_entry_safe(trav, tmp, &saved_frames->sf.list, list) + { + gf_time_fmt_tv(timestr, sizeof timestr, &trav->saved_at, gf_timefmt_FT); + + if (!trav->rpcreq || !trav->rpcreq->prog) + continue; + + gf_log_callingfn( + trav->rpcreq->conn->name, GF_LOG_ERROR, + "forced unwinding frame type(%s) op(%s(%d)) " + "called at %s (xid=0x%x)", + trav->rpcreq->prog->progname, + ((trav->rpcreq->prog->procnames) + ? trav->rpcreq->prog->procnames[trav->rpcreq->procnum] + : "--"), + trav->rpcreq->procnum, timestr, trav->rpcreq->xid); + saved_frames->count--; + + trav->rpcreq->rpc_status = -1; + trav->rpcreq->cbkfn(trav->rpcreq, NULL, 0, trav->frame); + + rpc_clnt_reply_deinit(trav->rpcreq, + trav->rpcreq->conn->rpc_clnt->reqpool); + + list_del_init(&trav->list); + mem_put(trav); + } +} + +void +saved_frames_destroy(struct saved_frames *frames) +{ + if (!frames) + return; + + saved_frames_unwind(frames); + + GF_FREE(frames); +} + +void +rpc_clnt_reconnect(void *conn_ptr) +{ + rpc_transport_t *trans = NULL; + rpc_clnt_connection_t *conn = NULL; + struct timespec ts = {0, 0}; + struct rpc_clnt *clnt = NULL; + gf_boolean_t need_unref = _gf_false; + gf_boolean_t canceled_unref = _gf_false; + + conn = conn_ptr; + clnt = conn->rpc_clnt; + pthread_mutex_lock(&conn->lock); + { + trans = conn->trans; + if (!trans) + goto out_unlock; + + if (conn->reconnect) { + if (!gf_timer_call_cancel(clnt->ctx, conn->reconnect)) + canceled_unref = _gf_true; + } + conn->reconnect = 0; + + if ((conn->connected == 0) && !clnt->disabled) { + ts.tv_sec = 3; + ts.tv_nsec = 0; + + gf_log(conn->name, GF_LOG_TRACE, "attempting reconnect"); + (void)rpc_transport_connect(trans, conn->config.remote_port); + rpc_clnt_ref(clnt); + conn->reconnect = gf_timer_call_after(clnt->ctx, ts, + rpc_clnt_reconnect, conn); + if (!conn->reconnect) { + need_unref = _gf_true; + gf_log(conn->name, GF_LOG_ERROR, + "Error adding to timer event queue"); + } + } else { + gf_log(conn->name, GF_LOG_TRACE, "breaking reconnect chain"); + } + } +out_unlock: + pthread_mutex_unlock(&conn->lock); + + rpc_clnt_unref(clnt); + if (need_unref) + rpc_clnt_unref(clnt); + if (canceled_unref) + rpc_clnt_unref(clnt); + return; +} + +int +rpc_clnt_fill_request_info(struct rpc_clnt *clnt, rpc_request_info_t *info) +{ + struct saved_frame saved_frame; + int ret = -1; + + pthread_mutex_lock(&clnt->conn.lock); + { + ret = __saved_frame_copy(clnt->conn.saved_frames, info->xid, + &saved_frame); + } + pthread_mutex_unlock(&clnt->conn.lock); + + if (ret == -1) { + gf_log(clnt->conn.name, GF_LOG_CRITICAL, + "cannot lookup the saved " + "frame corresponding to xid (%d)", + info->xid); + goto out; + } + + info->prognum = saved_frame.rpcreq->prog->prognum; + info->procnum = saved_frame.rpcreq->procnum; + info->progver = saved_frame.rpcreq->prog->progver; + info->rpc_req = saved_frame.rpcreq; + info->rsp = saved_frame.rsp; + + ret = 0; +out: + return ret; +} + +int +rpc_clnt_reconnect_cleanup(rpc_clnt_connection_t *conn) +{ + struct rpc_clnt *clnt = NULL; + int ret = 0; + gf_boolean_t reconnect_unref = _gf_false; + + if (!conn) { + goto out; + } + + clnt = conn->rpc_clnt; + + pthread_mutex_lock(&conn->lock); + { + if (conn->reconnect) { + ret = gf_timer_call_cancel(clnt->ctx, conn->reconnect); + if (!ret) { + reconnect_unref = _gf_true; + conn->cleanup_gen++; + } + conn->reconnect = NULL; + } + } + pthread_mutex_unlock(&conn->lock); + + if (reconnect_unref) + rpc_clnt_unref(clnt); + +out: + return 0; +} + +/* + * client_protocol_cleanup - cleanup function + * @trans: transport object + * + */ +int +rpc_clnt_connection_cleanup(rpc_clnt_connection_t *conn) +{ + struct saved_frames *saved_frames = NULL; + struct rpc_clnt *clnt = NULL; + int unref = 0; + int ret = 0; + gf_boolean_t timer_unref = _gf_false; + gf_boolean_t reconnect_unref = _gf_false; + + if (!conn) { + goto out; + } + + clnt = conn->rpc_clnt; + + pthread_mutex_lock(&conn->lock); + { + saved_frames = conn->saved_frames; + conn->saved_frames = saved_frames_new(); + + /* bailout logic cleanup */ + if (conn->timer) { + ret = gf_timer_call_cancel(clnt->ctx, conn->timer); + if (!ret) + timer_unref = _gf_true; + conn->timer = NULL; + } + if (conn->reconnect) { + ret = gf_timer_call_cancel(clnt->ctx, conn->reconnect); + if (!ret) + reconnect_unref = _gf_true; + conn->reconnect = NULL; + } + + conn->connected = 0; + conn->disconnected = 1; + + unref = rpc_clnt_remove_ping_timer_locked(clnt); + /*reset rpc msgs stats*/ + conn->pingcnt = 0; + conn->msgcnt = 0; + conn->cleanup_gen++; + } + pthread_mutex_unlock(&conn->lock); + + saved_frames_destroy(saved_frames); + if (unref) + rpc_clnt_unref(clnt); + + if (timer_unref) + rpc_clnt_unref(clnt); + + if (reconnect_unref) + rpc_clnt_unref(clnt); +out: + return 0; +} + +/* + * lookup_frame - lookup call frame corresponding to a given callid + * @trans: transport object + * @callid: call id of the frame + * + * not for external reference + */ + +static struct saved_frame * +lookup_frame(rpc_clnt_connection_t *conn, int64_t callid) +{ + struct saved_frame *frame = NULL; + + pthread_mutex_lock(&conn->lock); + { + frame = __saved_frame_get(conn->saved_frames, callid); + } + pthread_mutex_unlock(&conn->lock); + + return frame; +} + +int +rpc_clnt_reply_fill(rpc_transport_pollin_t *msg, rpc_clnt_connection_t *conn, + struct rpc_msg *replymsg, struct iovec progmsg, + struct rpc_req *req, struct saved_frame *saved_frame) +{ + int ret = -1; + + if ((!conn) || (!replymsg) || (!req) || (!saved_frame) || (!msg)) { + goto out; + } + + req->rpc_status = 0; + if ((rpc_reply_status(replymsg) == MSG_DENIED) || + (rpc_accepted_reply_status(replymsg) != SUCCESS)) { + req->rpc_status = -1; + } + + req->rsp[0] = progmsg; + req->rsp_iobref = iobref_ref(msg->iobref); + + if (msg->vectored) { + req->rsp[1] = msg->vector[1]; + req->rspcnt = 2; + } else { + req->rspcnt = 1; + } + + /* By this time, the data bytes for the auth scheme would have already + * been copied into the required sections of the req structure, + * we just need to fill in the meta-data about it now. + */ + if (req->rpc_status == 0) { + /* + * req->verf.flavour = rpc_reply_verf_flavour (replymsg); + * req->verf.datalen = rpc_reply_verf_len (replymsg); + */ + } + + ret = 0; + +out: + return ret; +} + +void +rpc_clnt_reply_deinit(struct rpc_req *req, struct mem_pool *pool) +{ + if (!req) { + goto out; + } + + if (req->rsp_iobref) { + iobref_unref(req->rsp_iobref); + } + + mem_put(req); +out: + return; +} + +/* TODO: use mem-pool for allocating requests */ +int +rpc_clnt_reply_init(rpc_clnt_connection_t *conn, rpc_transport_pollin_t *msg, + struct rpc_req *req, struct saved_frame *saved_frame) +{ + char *msgbuf = NULL; + struct rpc_msg rpcmsg; + struct iovec progmsg; /* RPC Program payload */ + size_t msglen = 0; + int ret = -1; + + msgbuf = msg->vector[0].iov_base; + msglen = msg->vector[0].iov_len; + + ret = xdr_to_rpc_reply(msgbuf, msglen, &rpcmsg, &progmsg, + req->verf.authdata); + if (ret != 0) { + gf_log(conn->name, GF_LOG_WARNING, "RPC reply decoding failed"); + goto out; + } + + ret = rpc_clnt_reply_fill(msg, conn, &rpcmsg, progmsg, req, saved_frame); + if (ret != 0) { + goto out; + } + + gf_log(conn->name, GF_LOG_TRACE, + "received rpc message (RPC XID: 0x%x" + " Program: %s, ProgVers: %d, Proc: %d) from rpc-transport (%s)", + saved_frame->rpcreq->xid, saved_frame->rpcreq->prog->progname, + saved_frame->rpcreq->prog->progver, saved_frame->rpcreq->procnum, + conn->name); + +out: + if (ret != 0) { + req->rpc_status = -1; + } + + return ret; +} + +int +rpc_clnt_handle_cbk(struct rpc_clnt *clnt, rpc_transport_pollin_t *msg) +{ + char *msgbuf = NULL; + rpcclnt_cb_program_t *program = NULL; + struct rpc_msg rpcmsg; + struct iovec progmsg; /* RPC Program payload */ + size_t msglen = 0; + int found = 0; + int ret = -1; + int procnum = 0; + + msgbuf = msg->vector[0].iov_base; + msglen = msg->vector[0].iov_len; + + clnt = rpc_clnt_ref(clnt); + ret = xdr_to_rpc_call(msgbuf, msglen, &rpcmsg, &progmsg, NULL, NULL); + if (ret == -1) { + gf_log(clnt->conn.name, GF_LOG_WARNING, "RPC call decoding failed"); + goto out; + } + + gf_log(clnt->conn.name, GF_LOG_TRACE, + "receivd rpc message (XID: 0x%" GF_PRI_RPC_XID + ", " + "Ver: %" GF_PRI_RPC_VERSION ", Program: %" GF_PRI_RPC_PROG_ID + ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC + ") " + "from rpc-transport (%s)", + rpc_call_xid(&rpcmsg), rpc_call_rpcvers(&rpcmsg), + rpc_call_program(&rpcmsg), rpc_call_progver(&rpcmsg), + rpc_call_progproc(&rpcmsg), clnt->conn.name); + + procnum = rpc_call_progproc(&rpcmsg); + + pthread_mutex_lock(&clnt->lock); + { + list_for_each_entry(program, &clnt->programs, program) + { + if ((program->prognum == rpc_call_program(&rpcmsg)) && + (program->progver == rpc_call_progver(&rpcmsg))) { + found = 1; + break; + } + } + } + pthread_mutex_unlock(&clnt->lock); + + if (found && (procnum < program->numactors) && + (program->actors[procnum].actor)) { + program->actors[procnum].actor(clnt, program->mydata, &progmsg); + } + +out: + rpc_clnt_unref(clnt); + return ret; +} + +int +rpc_clnt_handle_reply(struct rpc_clnt *clnt, rpc_transport_pollin_t *pollin) +{ + rpc_clnt_connection_t *conn = NULL; + struct saved_frame *saved_frame = NULL; + int ret = -1; + struct rpc_req *req = NULL; + uint32_t xid = 0; + + clnt = rpc_clnt_ref(clnt); + conn = &clnt->conn; + + xid = ntoh32(*((uint32_t *)pollin->vector[0].iov_base)); + saved_frame = lookup_frame(conn, xid); + if (saved_frame == NULL) { + gf_log(conn->name, GF_LOG_ERROR, + "cannot lookup the saved frame for reply with xid (%u)", xid); + goto out; + } + + req = saved_frame->rpcreq; + if (req == NULL) { + gf_log(conn->name, GF_LOG_ERROR, "no request with frame for xid (%u)", + xid); + goto out; + } + + ret = rpc_clnt_reply_init(conn, pollin, req, saved_frame); + if (ret != 0) { + req->rpc_status = -1; + gf_log(conn->name, GF_LOG_WARNING, "initialising rpc reply failed"); + } + + req->cbkfn(req, req->rsp, req->rspcnt, saved_frame->frame); + + if (req) { + rpc_clnt_reply_deinit(req, conn->rpc_clnt->reqpool); + } +out: + + if (saved_frame) { + mem_put(saved_frame); + } + + rpc_clnt_unref(clnt); + return ret; +} + +gf_boolean_t +is_rpc_clnt_disconnected(rpc_clnt_connection_t *conn) +{ + gf_boolean_t disconnected = _gf_true; + + if (!conn) + return disconnected; + + pthread_mutex_lock(&conn->lock); + { + disconnected = conn->disconnected; + } + pthread_mutex_unlock(&conn->lock); + + return disconnected; +} + +static void +rpc_clnt_destroy(struct rpc_clnt *rpc); + +#define RPC_THIS_SAVE(xl) \ + do { \ + old_THIS = THIS; \ + if (!old_THIS) \ + gf_log_callingfn("rpc", GF_LOG_CRITICAL, \ + "THIS is not initialised."); \ + THIS = xl; \ + } while (0) + +#define RPC_THIS_RESTORE (THIS = old_THIS) + +static int +rpc_clnt_handle_disconnect(struct rpc_clnt *clnt, rpc_clnt_connection_t *conn) +{ + struct timespec ts = { + 0, + }; + gf_boolean_t unref_clnt = _gf_false; + uint64_t pre_notify_gen = 0, post_notify_gen = 0; + + pthread_mutex_lock(&conn->lock); + { + pre_notify_gen = conn->cleanup_gen; + } + pthread_mutex_unlock(&conn->lock); + + if (clnt->notifyfn) + clnt->notifyfn(clnt, clnt->mydata, RPC_CLNT_DISCONNECT, NULL); + + pthread_mutex_lock(&conn->lock); + { + post_notify_gen = conn->cleanup_gen; + } + pthread_mutex_unlock(&conn->lock); + + if (pre_notify_gen == post_notify_gen) { + /* program didn't invoke cleanup, so rpc has to do it */ + rpc_clnt_connection_cleanup(conn); + } + + pthread_mutex_lock(&conn->lock); + { + if (!conn->rpc_clnt->disabled && (conn->reconnect == NULL)) { + ts.tv_sec = 3; + ts.tv_nsec = 0; + + rpc_clnt_ref(clnt); + conn->reconnect = gf_timer_call_after(clnt->ctx, ts, + rpc_clnt_reconnect, conn); + if (conn->reconnect == NULL) { + gf_log(conn->name, GF_LOG_WARNING, + "Cannot create rpc_clnt_reconnect timer"); + unref_clnt = _gf_true; + } + } + } + pthread_mutex_unlock(&conn->lock); + + if (unref_clnt) + rpc_clnt_unref(clnt); + + return 0; +} + +int +rpc_clnt_notify(rpc_transport_t *trans, void *mydata, + rpc_transport_event_t event, void *data, ...) +{ + rpc_clnt_connection_t *conn = NULL; + struct rpc_clnt *clnt = NULL; + int ret = -1; + rpc_request_info_t *req_info = NULL; + rpc_transport_pollin_t *pollin = NULL; + void *clnt_mydata = NULL; + DECLARE_OLD_THIS; + + conn = mydata; + if (conn == NULL) { + goto out; + } + clnt = conn->rpc_clnt; + if (!clnt) + goto out; + + RPC_THIS_SAVE(clnt->owner); + + switch (event) { + case RPC_TRANSPORT_DISCONNECT: { + rpc_clnt_handle_disconnect(clnt, conn); + /* The auth_value was being reset to AUTH_GLUSTERFS_v2. + * if (clnt->auth_value) + * clnt->auth_value = AUTH_GLUSTERFS_v2; + * It should not be reset here. The disconnect during + * portmap request can race with handshake. If handshake + * happens first and disconnect later, auth_value would set + * to default value and it never sets back to actual auth_value + * supported by server. But it's important to set to lower + * version supported in the case where the server downgrades. + * So moving this code to RPC_TRANSPORT_CONNECT. Note that + * CONNECT cannot race with handshake as by nature it is + * serialized with handhake. An handshake can happen only + * on a connected transport and hence its strictly serialized. + */ + break; + } + + case RPC_TRANSPORT_CLEANUP: + if (clnt->notifyfn) { + clnt_mydata = clnt->mydata; + clnt->mydata = NULL; + ret = clnt->notifyfn(clnt, clnt_mydata, RPC_CLNT_DESTROY, NULL); + if (ret < 0) { + gf_log(trans->name, GF_LOG_WARNING, + "client notify handler returned error " + "while handling RPC_CLNT_DESTROY"); + } + } + rpc_clnt_destroy(clnt); + ret = 0; + break; + + case RPC_TRANSPORT_MAP_XID_REQUEST: { + req_info = data; + ret = rpc_clnt_fill_request_info(clnt, req_info); + break; + } + + case RPC_TRANSPORT_MSG_RECEIVED: { + timespec_now_realtime(&conn->last_received); + + pollin = data; + if (pollin->is_reply) + ret = rpc_clnt_handle_reply(clnt, pollin); + else + ret = rpc_clnt_handle_cbk(clnt, pollin); + /* ret = clnt->notifyfn (clnt, clnt->mydata, RPC_CLNT_MSG, + * data); + */ + break; + } + + case RPC_TRANSPORT_MSG_SENT: { + timespec_now_realtime(&conn->last_sent); + ret = 0; + break; + } + + case RPC_TRANSPORT_CONNECT: { + pthread_mutex_lock(&conn->lock); + { + /* Every time there is a disconnection, processes + * should try to connect to 'glusterd' (ie, default + * port) or whichever port given as 'option remote-port' + * in volume file. */ + /* Below code makes sure the (re-)configured port lasts + * for just one successful attempt */ + conn->config.remote_port = 0; + conn->connected = 1; + conn->disconnected = 0; + pthread_cond_broadcast(&conn->cond); + } + pthread_mutex_unlock(&conn->lock); + + /* auth value should be set to lower version available + * and will be set to appropriate version supported by + * server after the handshake. + */ + if (clnt->auth_value) + clnt->auth_value = AUTH_GLUSTERFS_v2; + if (clnt->notifyfn) + ret = clnt->notifyfn(clnt, clnt->mydata, RPC_CLNT_CONNECT, + NULL); + + break; + } + + case RPC_TRANSPORT_ACCEPT: + /* only meaningful on a server, no need of handling this event + * in a client. + */ + ret = 0; + break; + + case RPC_TRANSPORT_EVENT_THREAD_DIED: + /* only meaningful on a server, no need of handling this event on a + * client */ + ret = 0; + break; + } + +out: + RPC_THIS_RESTORE; + return ret; +} + +static int +rpc_clnt_connection_init(struct rpc_clnt *clnt, glusterfs_ctx_t *ctx, + dict_t *options, char *name) +{ + int ret = -1; + rpc_clnt_connection_t *conn = NULL; + rpc_transport_t *trans = NULL; + + conn = &clnt->conn; + pthread_mutex_init(&clnt->conn.lock, NULL); + pthread_cond_init(&clnt->conn.cond, NULL); + + conn->name = gf_strdup(name); + if (!conn->name) { + ret = -1; + goto out; + } + + ret = dict_get_int32(options, "frame-timeout", &conn->frame_timeout); + if (ret >= 0) { + gf_log(name, GF_LOG_INFO, "setting frame-timeout to %d", + conn->frame_timeout); + } else { + gf_log(name, GF_LOG_DEBUG, "defaulting frame-timeout to 30mins"); + conn->frame_timeout = 1800; + } + conn->rpc_clnt = clnt; + + ret = dict_get_int32(options, "ping-timeout", &conn->ping_timeout); + if (ret >= 0) { + gf_log(name, GF_LOG_DEBUG, "setting ping-timeout to %d", + conn->ping_timeout); + } else { + /*TODO: Once the epoll thread model is fixed, + change the default ping-timeout to 30sec */ + gf_log(name, GF_LOG_DEBUG, "disable ping-timeout"); + conn->ping_timeout = 0; + } + + trans = rpc_transport_load(ctx, options, name); + if (!trans) { + gf_log(name, GF_LOG_WARNING, + "loading of new rpc-transport" + " failed"); + ret = -1; + goto out; + } + rpc_transport_ref(trans); + + pthread_mutex_lock(&conn->lock); + { + conn->trans = trans; + trans = NULL; + } + pthread_mutex_unlock(&conn->lock); + + ret = rpc_transport_register_notify(conn->trans, rpc_clnt_notify, conn); + if (ret == -1) { + gf_log(name, GF_LOG_WARNING, "registering notify failed"); + goto out; + } + + conn->saved_frames = saved_frames_new(); + if (!conn->saved_frames) { + gf_log(name, GF_LOG_WARNING, + "creation of saved_frames " + "failed"); + ret = -1; + goto out; + } + + ret = 0; + +out: + if (ret) { + pthread_mutex_lock(&conn->lock); + { + trans = conn->trans; + conn->trans = NULL; + } + pthread_mutex_unlock(&conn->lock); + if (trans) + rpc_transport_unref(trans); + // conn cleanup needs to be done since we might have failed to + // register notification. + rpc_clnt_connection_cleanup(conn); + } + return ret; +} + +struct rpc_clnt * +rpc_clnt_new(dict_t *options, xlator_t *owner, char *name, + uint32_t reqpool_size) +{ + int ret = -1; + struct rpc_clnt *rpc = NULL; + glusterfs_ctx_t *ctx = owner->ctx; + + rpc = GF_CALLOC(1, sizeof(*rpc), gf_common_mt_rpcclnt_t); + if (!rpc) { + goto out; + } + + pthread_mutex_init(&rpc->lock, NULL); + rpc->ctx = ctx; + rpc->owner = owner; + GF_ATOMIC_INIT(rpc->xid, 1); + + if (!reqpool_size) + reqpool_size = RPC_CLNT_DEFAULT_REQUEST_COUNT; + + rpc->reqpool = mem_pool_new(struct rpc_req, reqpool_size); + if (rpc->reqpool == NULL) { + pthread_mutex_destroy(&rpc->lock); + GF_FREE(rpc); + rpc = NULL; + goto out; + } + + rpc->saved_frames_pool = mem_pool_new(struct saved_frame, reqpool_size); + if (rpc->saved_frames_pool == NULL) { + pthread_mutex_destroy(&rpc->lock); + mem_pool_destroy(rpc->reqpool); + GF_FREE(rpc); + rpc = NULL; + goto out; + } + + ret = rpc_clnt_connection_init(rpc, ctx, options, name); + if (ret == -1) { + pthread_mutex_destroy(&rpc->lock); + mem_pool_destroy(rpc->reqpool); + mem_pool_destroy(rpc->saved_frames_pool); + GF_FREE(rpc); + rpc = NULL; + goto out; + } + + /* This is handled to make sure we have modularity in getting the + auth data changed */ + gf_boolean_t auth_null = dict_get_str_boolean(options, "auth-null", 0); + + rpc->auth_value = (auth_null) ? 0 : AUTH_GLUSTERFS_v2; + + rpc = rpc_clnt_ref(rpc); + INIT_LIST_HEAD(&rpc->programs); + +out: + return rpc; +} + +int +rpc_clnt_start(struct rpc_clnt *rpc) +{ + struct rpc_clnt_connection *conn = NULL; + + if (!rpc) + return -1; + + conn = &rpc->conn; + + pthread_mutex_lock(&conn->lock); + { + rpc->disabled = 0; + } + pthread_mutex_unlock(&conn->lock); + /* Corresponding unref will be either on successful timer cancel or last + * rpc_clnt_reconnect fire event. + */ + rpc_clnt_ref(rpc); + rpc_clnt_reconnect(conn); + + return 0; +} + +int +rpc_clnt_cleanup_and_start(struct rpc_clnt *rpc) +{ + struct rpc_clnt_connection *conn = NULL; + + if (!rpc) + return -1; + + conn = &rpc->conn; + + rpc_clnt_connection_cleanup(conn); + + pthread_mutex_lock(&conn->lock); + { + rpc->disabled = 0; + } + pthread_mutex_unlock(&conn->lock); + /* Corresponding unref will be either on successful timer cancel or last + * rpc_clnt_reconnect fire event. + */ + rpc_clnt_ref(rpc); + rpc_clnt_reconnect(conn); + + return 0; +} + +int +rpc_clnt_register_notify(struct rpc_clnt *rpc, rpc_clnt_notify_t fn, + void *mydata) +{ + rpc->mydata = mydata; + rpc->notifyfn = fn; + + return 0; +} + +/* used for GF_LOG_OCCASIONALLY() */ +static int gf_auth_max_groups_log = 0; + +static inline int +setup_glusterfs_auth_param_v3(call_frame_t *frame, auth_glusterfs_params_v3 *au, + int lk_owner_len, char *owner_data) +{ + int ret = -1; + unsigned int max_groups = 0; + int max_lkowner_len = 0; + + au->pid = frame->root->pid; + au->uid = frame->root->uid; + au->gid = frame->root->gid; + + au->flags = frame->root->flags; + au->ctime_sec = frame->root->ctime.tv_sec; + au->ctime_nsec = frame->root->ctime.tv_nsec; + + au->lk_owner.lk_owner_val = owner_data; + au->lk_owner.lk_owner_len = lk_owner_len; + au->groups.groups_val = frame->root->groups; + au->groups.groups_len = frame->root->ngrps; + + /* The number of groups and the size of lk_owner depend on oneother. + * We can truncate the groups, but should not touch the lk_owner. */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS(lk_owner_len, AUTH_GLUSTERFS_v3); + if (au->groups.groups_len > max_groups) { + GF_LOG_OCCASIONALLY(gf_auth_max_groups_log, "rpc-auth", GF_LOG_WARNING, + "truncating grouplist " + "from %d to %d", + au->groups.groups_len, max_groups); + + au->groups.groups_len = max_groups; + } + + max_lkowner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER(au->groups.groups_len, + AUTH_GLUSTERFS_v3); + if (lk_owner_len > max_lkowner_len) { + gf_log("rpc-clnt", GF_LOG_ERROR, + "lkowner field is too " + "big (%d), it does not fit in the rpc-header", + au->lk_owner.lk_owner_len); + errno = E2BIG; + goto out; + } + + ret = 0; +out: + return ret; +} + +static inline int +setup_glusterfs_auth_param_v2(call_frame_t *frame, auth_glusterfs_parms_v2 *au, + int lk_owner_len, char *owner_data) +{ + unsigned int max_groups = 0; + int max_lkowner_len = 0; + int ret = -1; + + au->pid = frame->root->pid; + au->uid = frame->root->uid; + au->gid = frame->root->gid; + + au->lk_owner.lk_owner_val = owner_data; + au->lk_owner.lk_owner_len = lk_owner_len; + au->groups.groups_val = frame->root->groups; + au->groups.groups_len = frame->root->ngrps; + + /* The number of groups and the size of lk_owner depend on oneother. + * We can truncate the groups, but should not touch the lk_owner. */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS(lk_owner_len, AUTH_GLUSTERFS_v2); + if (au->groups.groups_len > max_groups) { + GF_LOG_OCCASIONALLY(gf_auth_max_groups_log, "rpc-auth", GF_LOG_WARNING, + "truncating grouplist " + "from %d to %d", + au->groups.groups_len, max_groups); + + au->groups.groups_len = max_groups; + } + + max_lkowner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER(au->groups.groups_len, + AUTH_GLUSTERFS_v2); + if (lk_owner_len > max_lkowner_len) { + gf_log("rpc-auth", GF_LOG_ERROR, + "lkowner field is too " + "big (%d), it does not fit in the rpc-header", + au->lk_owner.lk_owner_len); + errno = E2BIG; + goto out; + } + + ret = 0; +out: + return ret; +} + +static ssize_t +xdr_serialize_glusterfs_auth(struct rpc_clnt *clnt, call_frame_t *frame, + char *dest) +{ + ssize_t ret = -1; + XDR xdr; + char owner[4] = { + 0, + }; + int32_t pid = 0; + char *lk_owner_data = NULL; + int lk_owner_len = 0; + + if ((!dest)) + return -1; + + xdrmem_create(&xdr, dest, GF_MAX_AUTH_BYTES, XDR_ENCODE); + + if (frame->root->lk_owner.len) { + lk_owner_data = frame->root->lk_owner.data; + lk_owner_len = frame->root->lk_owner.len; + } else { + pid = frame->root->pid; + owner[0] = (char)(pid & 0xff); + owner[1] = (char)((pid >> 8) & 0xff); + owner[2] = (char)((pid >> 16) & 0xff); + owner[3] = (char)((pid >> 24) & 0xff); + + lk_owner_data = owner; + lk_owner_len = 4; + } + + if (clnt->auth_value == AUTH_GLUSTERFS_v2) { + auth_glusterfs_parms_v2 au_v2 = { + 0, + }; + + ret = setup_glusterfs_auth_param_v2(frame, &au_v2, lk_owner_len, + lk_owner_data); + if (ret) + goto out; + if (!xdr_auth_glusterfs_parms_v2(&xdr, &au_v2)) { + gf_log(THIS->name, GF_LOG_WARNING, + "failed to encode auth glusterfs elements"); + ret = -1; + goto out; + } + } else if (clnt->auth_value == AUTH_GLUSTERFS_v3) { + auth_glusterfs_params_v3 au_v3 = { + 0, + }; + + ret = setup_glusterfs_auth_param_v3(frame, &au_v3, lk_owner_len, + lk_owner_data); + if (ret) + goto out; + + if (!xdr_auth_glusterfs_params_v3(&xdr, &au_v3)) { + gf_log(THIS->name, GF_LOG_WARNING, + "failed to encode auth glusterfs elements"); + ret = -1; + goto out; + } + } else { + gf_log(THIS->name, GF_LOG_WARNING, + "failed to encode auth glusterfs elements"); + ret = -1; + goto out; + } + + ret = (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)); + +out: + return ret; +} + +int +rpc_clnt_fill_request(struct rpc_clnt *clnt, int prognum, int progver, + int procnum, uint64_t xid, call_frame_t *fr, + struct rpc_msg *request, char *auth_data) +{ + int ret = -1; + + if (!request) { + goto out; + } + + memset(request, 0, sizeof(*request)); + + request->rm_xid = xid; + request->rm_direction = CALL; + + request->rm_call.cb_rpcvers = 2; + request->rm_call.cb_prog = prognum; + request->rm_call.cb_vers = progver; + request->rm_call.cb_proc = procnum; + + if (!clnt->auth_value) { + request->rm_call.cb_cred.oa_flavor = AUTH_NULL; + request->rm_call.cb_cred.oa_base = NULL; + request->rm_call.cb_cred.oa_length = 0; + } else { + ret = xdr_serialize_glusterfs_auth(clnt, fr, auth_data); + if (ret == -1) { + gf_log("rpc-clnt", GF_LOG_WARNING, + "cannot encode auth credentials"); + goto out; + } + + request->rm_call.cb_cred.oa_flavor = clnt->auth_value; + request->rm_call.cb_cred.oa_base = auth_data; + request->rm_call.cb_cred.oa_length = ret; + } + request->rm_call.cb_verf.oa_flavor = AUTH_NONE; + request->rm_call.cb_verf.oa_base = NULL; + request->rm_call.cb_verf.oa_length = 0; + + ret = 0; +out: + return ret; +} + +struct iovec +rpc_clnt_record_build_header(char *recordstart, size_t rlen, + struct rpc_msg *request, size_t payload) +{ + struct iovec requesthdr = { + 0, + }; + struct iovec txrecord = {0, 0}; + int ret = -1; + size_t fraglen = 0; + + ret = rpc_request_to_xdr(request, recordstart, rlen, &requesthdr); + if (ret == -1) { + gf_log("rpc-clnt", GF_LOG_DEBUG, "Failed to create RPC request"); + goto out; + } + + fraglen = payload + requesthdr.iov_len; + gf_log("rpc-clnt", GF_LOG_TRACE, + "Request fraglen %zu, payload: %zu, " + "rpc hdr: %zu", + fraglen, payload, requesthdr.iov_len); + + txrecord.iov_base = recordstart; + + /* Remember, this is only the vec for the RPC header and does not + * include the payload above. We needed the payload only to calculate + * the size of the full fragment. This size is sent in the fragment + * header. + */ + txrecord.iov_len = requesthdr.iov_len; + +out: + return txrecord; +} + +struct iobuf * +rpc_clnt_record_build_record(struct rpc_clnt *clnt, call_frame_t *fr, + int prognum, int progver, int procnum, + size_t hdrsize, uint64_t xid, struct iovec *recbuf) +{ + struct rpc_msg request = { + 0, + }; + struct iobuf *request_iob = NULL; + char *record = NULL; + struct iovec recordhdr = { + 0, + }; + size_t pagesize = 0; + int ret = -1; + size_t xdr_size = 0; + char auth_data[GF_MAX_AUTH_BYTES] = { + 0, + }; + + if ((!clnt) || (!recbuf)) { + goto out; + } + + /* Fill the rpc structure and XDR it into the buffer got above. */ + ret = rpc_clnt_fill_request(clnt, prognum, progver, procnum, xid, fr, + &request, auth_data); + + if (ret == -1) { + gf_log(clnt->conn.name, GF_LOG_WARNING, + "cannot build a rpc-request xid (%" PRIu64 ")", xid); + goto out; + } + + xdr_size = xdr_sizeof((xdrproc_t)xdr_callmsg, &request); + + /* First, try to get a pointer into the buffer which the RPC + * layer can use. + */ + request_iob = iobuf_get2(clnt->ctx->iobuf_pool, (xdr_size + hdrsize)); + if (!request_iob) { + goto out; + } + + pagesize = iobuf_pagesize(request_iob); + + record = iobuf_ptr(request_iob); /* Now we have it. */ + + recordhdr = rpc_clnt_record_build_header(record, pagesize, &request, + hdrsize); + + if (!recordhdr.iov_base) { + gf_log(clnt->conn.name, GF_LOG_ERROR, "Failed to build record header"); + iobuf_unref(request_iob); + request_iob = NULL; + recbuf->iov_base = NULL; + goto out; + } + + recbuf->iov_base = recordhdr.iov_base; + recbuf->iov_len = recordhdr.iov_len; + +out: + return request_iob; +} + +static inline struct iobuf * +rpc_clnt_record(struct rpc_clnt *clnt, call_frame_t *call_frame, + rpc_clnt_prog_t *prog, int procnum, size_t hdrlen, + struct iovec *rpchdr, uint64_t callid) +{ + if (!prog || !rpchdr || !call_frame) { + return NULL; + } + + return rpc_clnt_record_build_record(clnt, call_frame, prog->prognum, + prog->progver, procnum, hdrlen, callid, + rpchdr); +} + +int +rpcclnt_cbk_program_register(struct rpc_clnt *clnt, + rpcclnt_cb_program_t *program, void *mydata) +{ + int ret = -1; + char already_registered = 0; + rpcclnt_cb_program_t *tmp = NULL; + + if (!clnt) + goto out; + + if (program->actors == NULL) + goto out; + + pthread_mutex_lock(&clnt->lock); + { + list_for_each_entry(tmp, &clnt->programs, program) + { + if ((program->prognum == tmp->prognum) && + (program->progver == tmp->progver)) { + already_registered = 1; + break; + } + } + } + pthread_mutex_unlock(&clnt->lock); + + if (already_registered) { + gf_log_callingfn(clnt->conn.name, GF_LOG_DEBUG, "already registered"); + ret = 0; + goto out; + } + + tmp = GF_MALLOC(sizeof(*tmp), gf_common_mt_rpcclnt_cb_program_t); + if (tmp == NULL) { + goto out; + } + + memcpy(tmp, program, sizeof(*tmp)); + INIT_LIST_HEAD(&tmp->program); + + tmp->mydata = mydata; + + pthread_mutex_lock(&clnt->lock); + { + list_add_tail(&tmp->program, &clnt->programs); + } + pthread_mutex_unlock(&clnt->lock); + + ret = 0; + gf_log(clnt->conn.name, GF_LOG_DEBUG, + "New program registered: %s, Num: %d, Ver: %d", program->progname, + program->prognum, program->progver); + +out: + if (ret == -1 && clnt) { + gf_log(clnt->conn.name, GF_LOG_ERROR, + "Program registration failed:" + " %s, Num: %d, Ver: %d", + program->progname, program->prognum, program->progver); + } + + return ret; +} + +int +rpc_clnt_submit(struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, int procnum, + fop_cbk_fn_t cbkfn, struct iovec *proghdr, int proghdrcount, + struct iovec *progpayload, int progpayloadcount, + struct iobref *iobref, void *frame, struct iovec *rsphdr, + int rsphdr_count, struct iovec *rsp_payload, + int rsp_payload_count, struct iobref *rsp_iobref) +{ + rpc_clnt_connection_t *conn = NULL; + struct iobuf *request_iob = NULL; + struct iovec rpchdr = { + 0, + }; + struct rpc_req *rpcreq = NULL; + rpc_transport_req_t req; + int ret = -1; + int proglen = 0; + char new_iobref = 0; + uint64_t callid = 0; + gf_boolean_t need_unref = _gf_false; + call_frame_t *cframe = frame; + + if (!rpc || !prog || !frame) { + goto out; + } + + conn = &rpc->conn; + + rpcreq = mem_get(rpc->reqpool); + if (rpcreq == NULL) { + goto out; + } + + memset(rpcreq, 0, sizeof(*rpcreq)); + memset(&req, 0, sizeof(req)); + + if (!iobref) { + iobref = iobref_new(); + if (!iobref) { + goto out; + } + + new_iobref = 1; + } + + callid = GF_ATOMIC_INC(rpc->xid); + + rpcreq->prog = prog; + rpcreq->procnum = procnum; + rpcreq->conn = conn; + rpcreq->xid = callid; + rpcreq->cbkfn = cbkfn; + + ret = -1; + + if (proghdr) { + proglen += iov_length(proghdr, proghdrcount); + } + + request_iob = rpc_clnt_record(rpc, frame, prog, procnum, proglen, &rpchdr, + callid); + if (!request_iob) { + gf_log(conn->name, GF_LOG_WARNING, "cannot build rpc-record"); + goto out; + } + + iobref_add(iobref, request_iob); + + req.msg.rpchdr = &rpchdr; + req.msg.rpchdrcount = 1; + req.msg.proghdr = proghdr; + req.msg.proghdrcount = proghdrcount; + req.msg.progpayload = progpayload; + req.msg.progpayloadcount = progpayloadcount; + req.msg.iobref = iobref; + + req.rsp.rsphdr = rsphdr; + req.rsp.rsphdr_count = rsphdr_count; + req.rsp.rsp_payload = rsp_payload; + req.rsp.rsp_payload_count = rsp_payload_count; + req.rsp.rsp_iobref = rsp_iobref; + req.rpc_req = rpcreq; + + pthread_mutex_lock(&conn->lock); + { + if (conn->connected == 0) { + if (rpc->disabled) + goto unlock; + ret = rpc_transport_connect(conn->trans, conn->config.remote_port); + if (ret < 0) { + gf_log(conn->name, + (errno == EINPROGRESS) ? GF_LOG_DEBUG : GF_LOG_WARNING, + "error returned while attempting to " + "connect to host:%s, port:%d", + conn->config.remote_host, conn->config.remote_port); + goto unlock; + } + } + + ret = rpc_transport_submit_request(conn->trans, &req); + if (ret == -1) { + gf_log(conn->name, GF_LOG_WARNING, + "failed to submit rpc-request " + "(unique: %" PRIu64 + ", XID: 0x%x Program: %s, " + "ProgVers: %d, Proc: %d) to rpc-transport (%s)", + cframe->root->unique, rpcreq->xid, rpcreq->prog->progname, + rpcreq->prog->progver, rpcreq->procnum, conn->name); + } else if ((ret >= 0) && frame) { + /* Save the frame in queue */ + __save_frame(rpc, frame, rpcreq); + + /* A ref on rpc-clnt object is taken while registering + * call_bail to timer in __save_frame. If it fails to + * register, it needs an unref and should happen outside + * conn->lock which otherwise leads to deadlocks */ + if (conn->timer == NULL) + need_unref = _gf_true; + + conn->msgcnt++; + + gf_log("rpc-clnt", GF_LOG_TRACE, + "submitted request " + "(unique: %" PRIu64 + ", XID: 0x%x, Program: %s, " + "ProgVers: %d, Proc: %d) to rpc-transport (%s)", + cframe->root->unique, rpcreq->xid, rpcreq->prog->progname, + rpcreq->prog->progver, rpcreq->procnum, conn->name); + } + } +unlock: + pthread_mutex_unlock(&conn->lock); + + if (need_unref) + rpc_clnt_unref(rpc); + + if (ret == -1) { + goto out; + } + + rpc_clnt_check_and_start_ping(rpc); + ret = 0; + +out: + if (request_iob) { + iobuf_unref(request_iob); + } + + if (new_iobref && iobref) { + iobref_unref(iobref); + } + + if (frame && (ret == -1)) { + if (rpcreq) { + rpcreq->rpc_status = -1; + cbkfn(rpcreq, NULL, 0, frame); + mem_put(rpcreq); + } + } + return ret; +} + +struct rpc_clnt * +rpc_clnt_ref(struct rpc_clnt *rpc) +{ + if (!rpc) + return NULL; + + GF_ATOMIC_INC(rpc->refcount); + return rpc; +} + +static void +rpc_clnt_trigger_destroy(struct rpc_clnt *rpc) +{ + rpc_clnt_connection_t *conn = NULL; + rpc_transport_t *trans = NULL; + + if (!rpc) + return; + + /* reading conn->trans outside conn->lock is OK, since this is the last + * ref*/ + conn = &rpc->conn; + trans = conn->trans; + rpc_clnt_disable(rpc); + + /* This is to account for rpc_clnt_disable that might have been called + * before rpc_clnt_unref */ + if (trans) { + /* set conn->trans to NULL before rpc_transport_unref + * as rpc_transport_unref can potentially free conn + */ + conn->trans = NULL; + rpc_transport_unref(trans); + } +} + +static void +rpc_clnt_destroy(struct rpc_clnt *rpc) +{ + rpcclnt_cb_program_t *program = NULL; + rpcclnt_cb_program_t *tmp = NULL; + struct saved_frames *saved_frames = NULL; + rpc_clnt_connection_t *conn = NULL; + + if (!rpc) + return; + + conn = &rpc->conn; + GF_FREE(rpc->conn.name); + /* Access saved_frames in critical-section to avoid + crash in rpc_clnt_connection_cleanup at the time + of destroying saved frames + */ + pthread_mutex_lock(&conn->lock); + { + saved_frames = conn->saved_frames; + conn->saved_frames = NULL; + } + pthread_mutex_unlock(&conn->lock); + + saved_frames_destroy(saved_frames); + pthread_mutex_destroy(&rpc->lock); + pthread_mutex_destroy(&rpc->conn.lock); + pthread_cond_destroy(&rpc->conn.cond); + + /* mem-pool should be destroyed, otherwise, + it will cause huge memory leaks */ + mem_pool_destroy(rpc->reqpool); + mem_pool_destroy(rpc->saved_frames_pool); + + list_for_each_entry_safe(program, tmp, &rpc->programs, program) + { + GF_FREE(program); + } + + GF_FREE(rpc); + return; +} + +struct rpc_clnt * +rpc_clnt_unref(struct rpc_clnt *rpc) +{ + int count = 0; + + if (!rpc) + return NULL; + + count = GF_ATOMIC_DEC(rpc->refcount); + + if (!count) { + rpc_clnt_trigger_destroy(rpc); + return NULL; + } + return rpc; +} + +int +rpc_clnt_disable(struct rpc_clnt *rpc) +{ + rpc_clnt_connection_t *conn = NULL; + rpc_transport_t *trans = NULL; + int unref = 0; + int ret = 0; + gf_boolean_t timer_unref = _gf_false; + gf_boolean_t reconnect_unref = _gf_false; + + if (!rpc) { + goto out; + } + + conn = &rpc->conn; + + pthread_mutex_lock(&conn->lock); + { + rpc->disabled = 1; + + if (conn->timer) { + ret = gf_timer_call_cancel(rpc->ctx, conn->timer); + /* If the event is not fired and it actually cancelled + * the timer, do the unref else registered call back + * function will take care of it. + */ + if (!ret) + timer_unref = _gf_true; + conn->timer = NULL; + } + + if (conn->reconnect) { + ret = gf_timer_call_cancel(rpc->ctx, conn->reconnect); + if (!ret) + reconnect_unref = _gf_true; + conn->reconnect = NULL; + } + conn->connected = 0; + + unref = rpc_clnt_remove_ping_timer_locked(rpc); + trans = conn->trans; + } + pthread_mutex_unlock(&conn->lock); + + ret = -1; + if (trans) { + ret = rpc_transport_disconnect(trans, _gf_true); + /* The auth_value was being reset to AUTH_GLUSTERFS_v2. + * if (clnt->auth_value) + * clnt->auth_value = AUTH_GLUSTERFS_v2; + * It should not be reset here. The disconnect during + * portmap request can race with handshake. If handshake + * happens first and disconnect later, auth_value would set + * to default value and it never sets back to actual auth_value + * supported by server. But it's important to set to lower + * version supported in the case where the server downgrades. + * So moving this code to RPC_TRANSPORT_CONNECT. Note that + * CONNECT cannot race with handshake as by nature it is + * serialized with handhake. An handshake can happen only + * on a connected transport and hence its strictly serialized. + */ + } + if (unref) + rpc_clnt_unref(rpc); + + if (timer_unref) + rpc_clnt_unref(rpc); + + if (reconnect_unref) + rpc_clnt_unref(rpc); + +out: + return ret; +} + +void +rpc_clnt_reconfig(struct rpc_clnt *rpc, struct rpc_clnt_config *config) +{ + if (config->ping_timeout) { + if (config->ping_timeout != rpc->conn.ping_timeout) + gf_log(rpc->conn.name, GF_LOG_INFO, + "changing ping timeout to %d (from %d)", + config->ping_timeout, rpc->conn.ping_timeout); + + pthread_mutex_lock(&rpc->conn.lock); + { + rpc->conn.ping_timeout = config->ping_timeout; + } + pthread_mutex_unlock(&rpc->conn.lock); + } + + if (config->rpc_timeout) { + if (config->rpc_timeout != rpc->conn.config.rpc_timeout) + gf_log(rpc->conn.name, GF_LOG_INFO, + "changing timeout to %d (from %d)", config->rpc_timeout, + rpc->conn.config.rpc_timeout); + rpc->conn.config.rpc_timeout = config->rpc_timeout; + } + + if (config->remote_port) { + if (config->remote_port != rpc->conn.config.remote_port) + gf_log(rpc->conn.name, GF_LOG_INFO, "changing port to %d (from %d)", + config->remote_port, rpc->conn.config.remote_port); + + rpc->conn.config.remote_port = config->remote_port; + } + + if (config->remote_host) { + if (rpc->conn.config.remote_host) { + if (strcmp(rpc->conn.config.remote_host, config->remote_host)) + gf_log(rpc->conn.name, GF_LOG_INFO, + "changing hostname to %s (from %s)", config->remote_host, + rpc->conn.config.remote_host); + GF_FREE(rpc->conn.config.remote_host); + } else { + gf_log(rpc->conn.name, GF_LOG_INFO, "setting hostname to %s", + config->remote_host); + } + + rpc->conn.config.remote_host = gf_strdup(config->remote_host); + } +} diff --git a/rpc/rpc-lib/src/rpc-clnt.h b/rpc/rpc-lib/src/rpc-clnt.h new file mode 100644 index 00000000000..2945265200b --- /dev/null +++ b/rpc/rpc-lib/src/rpc-clnt.h @@ -0,0 +1,260 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef __RPC_CLNT_H +#define __RPC_CLNT_H + +#include <glusterfs/stack.h> +#include "rpc-transport.h" +#include <glusterfs/timer.h> +#include "xdr-common.h" +#include "glusterfs3.h" + +typedef enum { + RPC_CLNT_CONNECT, + RPC_CLNT_DISCONNECT, + RPC_CLNT_PING, + RPC_CLNT_MSG, + RPC_CLNT_DESTROY +} rpc_clnt_event_t; + +#define SFRAME_GET_PROGNUM(sframe) (sframe->rpcreq->prog->prognum) +#define SFRAME_GET_PROGVER(sframe) (sframe->rpcreq->prog->progver) +#define SFRAME_GET_PROCNUM(sframe) (sframe->rpcreq->procnum) + +struct rpc_req; +struct rpc_clnt; +struct rpc_clnt_config; +struct rpc_clnt_program; + +typedef int (*rpc_clnt_notify_t)(struct rpc_clnt *rpc, void *mydata, + rpc_clnt_event_t fn, void *data); + +typedef int (*fop_cbk_fn_t)(struct rpc_req *req, struct iovec *iov, int count, + void *myframe); + +typedef int (*clnt_fn_t)(call_frame_t *fr, xlator_t *xl, void *args); + +struct saved_frame { + union { + struct list_head list; + struct { + struct saved_frame *frame_next; + struct saved_frame *frame_prev; + }; + }; + void *capital_this; + void *frame; + struct rpc_req *rpcreq; + struct timeval saved_at; + rpc_transport_rsp_t rsp; +}; + +struct saved_frames { + int64_t count; + struct saved_frame sf; + struct saved_frame lk_sf; +}; + +/* Initialized by procnum */ +typedef struct rpc_clnt_procedure { + char *procname; + clnt_fn_t fn; +} rpc_clnt_procedure_t; + +typedef struct rpc_clnt_program { + char *progname; + int prognum; + int progver; + rpc_clnt_procedure_t *proctable; + char **procnames; + int numproc; +} rpc_clnt_prog_t; + +typedef int (*rpcclnt_cb_fn)(struct rpc_clnt *rpc, void *mydata, void *data); + +/* The descriptor for each procedure/actor that runs + * over the RPC service. + */ +typedef struct rpcclnt_actor_desc { + char procname[32]; + rpcclnt_cb_fn actor; + int procnum; +} rpcclnt_cb_actor_t; + +/* Describes a program and its version along with the function pointers + * required to handle the procedures/actors of each program/version. + * Never changed ever by any thread so no need for a lock. + */ +typedef struct rpcclnt_cb_program { + char progname[32]; + int prognum; + int progver; + rpcclnt_cb_actor_t *actors; /* All procedure handlers */ + /* Program specific state handed to actors */ + void *private; + + /* list member to link to list of registered services with rpc_clnt */ + struct list_head program; + + /* Needed for passing back in cb_actor */ + void *mydata; + int numactors; /* Num actors in actor array */ + +} rpcclnt_cb_program_t; + +typedef struct rpc_auth_data { + int flavour; + int datalen; + char authdata[GF_MAX_AUTH_BYTES]; +} rpc_auth_data_t; + +struct rpc_clnt_config { + int rpc_timeout; + int remote_port; + char *remote_host; + int ping_timeout; +}; + +#define rpc_auth_flavour(au) ((au).flavour) + +struct rpc_clnt_connection { + pthread_mutex_t lock; + pthread_cond_t cond; + rpc_transport_t *trans; + struct rpc_clnt_config config; + gf_timer_t *reconnect; + gf_timer_t *timer; + gf_timer_t *ping_timer; + struct rpc_clnt *rpc_clnt; + struct saved_frames *saved_frames; + struct timespec last_sent; + struct timespec last_received; + uint64_t pingcnt; + uint64_t msgcnt; + uint64_t cleanup_gen; + char *name; + int32_t ping_started; + int32_t frame_timeout; + int32_t ping_timeout; + gf_boolean_t disconnected; + char connected; +}; +typedef struct rpc_clnt_connection rpc_clnt_connection_t; + +struct rpc_req { + rpc_clnt_connection_t *conn; + struct iovec req[2]; + struct iobref *req_iobref; + struct iovec rsp[2]; + int rspcnt; + int reqcnt; + struct iobref *rsp_iobref; + rpc_clnt_prog_t *prog; + rpc_auth_data_t verf; + fop_cbk_fn_t cbkfn; + void *conn_private; + int procnum; + int rpc_status; + uint32_t xid; +}; + +typedef struct rpc_clnt { + pthread_mutex_t lock; + rpc_clnt_notify_t notifyfn; + rpc_clnt_connection_t conn; + void *mydata; + gf_atomic_t xid; + + /* list of cb programs registered with rpc-clnt */ + struct list_head programs; + + /* Memory pool for rpc_req_t */ + struct mem_pool *reqpool; + + struct mem_pool *saved_frames_pool; + + glusterfs_ctx_t *ctx; + gf_atomic_t refcount; + xlator_t *owner; + int auth_value; + char disabled; +} rpc_clnt_t; + +struct rpc_clnt * +rpc_clnt_new(dict_t *options, xlator_t *owner, char *name, + uint32_t reqpool_size); + +int +rpc_clnt_start(struct rpc_clnt *rpc); + +int +rpc_clnt_cleanup_and_start(struct rpc_clnt *rpc); + +int +rpc_clnt_register_notify(struct rpc_clnt *rpc, rpc_clnt_notify_t fn, + void *mydata); + +/* Some preconditions related to vectors holding responses. + * @rsphdr: should contain pointer to buffer which can hold response header + * and length of the program header. In case of procedures whose + * respnose size is not bounded (eg., glusterfs lookup), the length + * should be equal to size of buffer. + * @rsp_payload: should contain pointer and length of the bu + * + * 1. Both @rsp_hdr and @rsp_payload are optional. + * 2. The user of rpc_clnt_submit, if wants response hdr and payload in its own + * buffers, then it has to populate @rsphdr and @rsp_payload. + * 3. when @rsp_payload is not NULL, @rsphdr should + * also be filled with pointer to buffer to hold header and length + * of the header. + */ + +int +rpc_clnt_submit(struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, int procnum, + fop_cbk_fn_t cbkfn, struct iovec *proghdr, int proghdrcount, + struct iovec *progpayload, int progpayloadcount, + struct iobref *iobref, void *frame, struct iovec *rsphdr, + int rsphdr_count, struct iovec *rsp_payload, + int rsp_payload_count, struct iobref *rsp_iobref); + +struct rpc_clnt * +rpc_clnt_ref(struct rpc_clnt *rpc); + +struct rpc_clnt * +rpc_clnt_unref(struct rpc_clnt *rpc); + +int +rpc_clnt_connection_cleanup(rpc_clnt_connection_t *conn); +int +rpc_clnt_reconnect_cleanup(rpc_clnt_connection_t *conn); +gf_boolean_t +is_rpc_clnt_disconnected(rpc_clnt_connection_t *conn); + +void +rpc_clnt_reconnect(void *trans_ptr); + +void +rpc_clnt_reconfig(struct rpc_clnt *rpc, struct rpc_clnt_config *config); + +/* All users of RPC services should use this API to register their + * procedure handlers. + */ +int +rpcclnt_cbk_program_register(struct rpc_clnt *svc, + rpcclnt_cb_program_t *program, void *mydata); + +int +rpc_clnt_disable(struct rpc_clnt *rpc); + +int +rpc_clnt_mgmt_pmap_signout(glusterfs_ctx_t *ctx, char *brick_name); + +#endif /* !_RPC_CLNT_H */ diff --git a/rpc/rpc-lib/src/rpc-drc.c b/rpc/rpc-lib/src/rpc-drc.c new file mode 100644 index 00000000000..de8dc630626 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.c @@ -0,0 +1,855 @@ +/* + Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#ifndef RPC_DRC_H +#include "rpc-drc.h" +#endif +#include <glusterfs/locking.h> +#include <glusterfs/statedump.h> +#include <glusterfs/mem-pool.h> + +#include <netinet/in.h> +#include <unistd.h> + +/** + * rpcsvc_drc_op_destroy - Destroys the cached reply + * + * @param drc - the main drc structure + * @param reply - the cached reply to destroy + * @return NULL if reply is destroyed, reply otherwise + */ +static drc_cached_op_t * +rpcsvc_drc_op_destroy(rpcsvc_drc_globals_t *drc, drc_cached_op_t *reply) +{ + GF_ASSERT(drc); + GF_ASSERT(reply); + + if (reply->state == DRC_OP_IN_TRANSIT) + return reply; + + iobref_unref(reply->msg.iobref); + if (reply->msg.rpchdr) + GF_FREE(reply->msg.rpchdr); + if (reply->msg.proghdr) + GF_FREE(reply->msg.proghdr); + if (reply->msg.progpayload) + GF_FREE(reply->msg.progpayload); + + list_del(&reply->global_list); + reply->client->op_count--; + drc->op_count--; + mem_put(reply); + reply = NULL; + + return reply; +} + +/** + * rpcsvc_drc_op_rb_unref - This function is used in rb tree cleanup only + * + * @param reply - the cached reply to unref + * @param drc - the main drc structure + * @return void + */ +static void +rpcsvc_drc_rb_op_destroy(void *reply, void *drc) +{ + rpcsvc_drc_op_destroy(drc, (drc_cached_op_t *)reply); +} + +/** + * rpcsvc_remove_drc_client - Cleanup the drc client + * + * @param client - the drc client to be removed + * @return void + */ +static void +rpcsvc_remove_drc_client(drc_client_t *client) +{ + rb_destroy(client->rbtree, rpcsvc_drc_rb_op_destroy); + list_del(&client->client_list); + GF_FREE(client); +} + +/** + * rpcsvc_client_lookup - Given a sockaddr_storage, find the client if it exists + * + * @param drc - the main drc structure + * @param sockaddr - the network address of the client to be looked up + * @return drc client if it exists, NULL otherwise + */ +static drc_client_t * +rpcsvc_client_lookup(rpcsvc_drc_globals_t *drc, + struct sockaddr_storage *sockaddr) +{ + drc_client_t *client = NULL; + + GF_ASSERT(drc); + GF_ASSERT(sockaddr); + + if (list_empty(&drc->clients_head)) + return NULL; + + list_for_each_entry(client, &drc->clients_head, client_list) + { + if (gf_sock_union_equal_addr(&client->sock_union, + (union gf_sock_union *)sockaddr)) + return client; + } + + return NULL; +} + +/** + * drc_compare_reqs - Used by rbtree to determine if incoming req matches with + * an existing node(cached reply) in rbtree + * + * @param item - pointer to the incoming req + * @param rb_node_data - pointer to an rbtree node (cached reply) + * @param param - drc pointer - unused here, but used in *op_destroy + * @return 0 if req matches reply, else (req->xid - reply->xid) + */ +int +drc_compare_reqs(const void *item, const void *rb_node_data, void *param) +{ + int ret = -1; + drc_cached_op_t *req = NULL; + drc_cached_op_t *reply = NULL; + + GF_ASSERT(item); + GF_ASSERT(rb_node_data); + GF_ASSERT(param); + + req = (drc_cached_op_t *)item; + reply = (drc_cached_op_t *)rb_node_data; + + ret = req->xid - reply->xid; + if (ret != 0) + return ret; + + if (req->prognum == reply->prognum && req->procnum == reply->procnum && + req->progversion == reply->progversion) + return 0; + + return 1; +} + +/** + * drc_init_client_cache - initialize a drc client and its rb tree + * + * @param drc - the main drc structure + * @param client - the drc client to be initialized + * @return 0 on success, -1 on failure + */ +static int +drc_init_client_cache(rpcsvc_drc_globals_t *drc, drc_client_t *client) +{ + GF_ASSERT(drc); + GF_ASSERT(client); + + client->rbtree = rb_create(drc_compare_reqs, drc, NULL); + if (!client->rbtree) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "rb tree creation failed"); + return -1; + } + + return 0; +} + +/** + * rpcsvc_get_drc_client - find the drc client with given sockaddr, else + * allocate and initialize a new drc client + * + * @param drc - the main drc structure + * @param sockaddr - network address of client + * @return drc client on success, NULL on failure + */ +static drc_client_t * +rpcsvc_get_drc_client(rpcsvc_drc_globals_t *drc, + struct sockaddr_storage *sockaddr) +{ + drc_client_t *client = NULL; + + GF_ASSERT(drc); + GF_ASSERT(sockaddr); + + client = rpcsvc_client_lookup(drc, sockaddr); + if (client) + goto out; + + /* if lookup fails, allocate cache for the new client */ + client = GF_CALLOC(1, sizeof(drc_client_t), gf_common_mt_drc_client_t); + if (!client) + goto out; + + GF_ATOMIC_INIT(client->ref, 0); + client->sock_union = (union gf_sock_union) * sockaddr; + client->op_count = 0; + INIT_LIST_HEAD(&client->client_list); + + if (drc_init_client_cache(drc, client)) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "initialization of drc client failed"); + GF_FREE(client); + client = NULL; + goto out; + } + drc->client_count++; + + list_add(&client->client_list, &drc->clients_head); + +out: + return client; +} + +/** + * rpcsvc_need_drc - Determine if a request needs DRC service + * + * @param req - incoming request + * @return 1 if DRC is needed for req, 0 otherwise + */ +int +rpcsvc_need_drc(rpcsvc_request_t *req) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT(req); + GF_ASSERT(req->svc); + + drc = req->svc->drc; + + if (!drc || drc->status == DRC_UNINITIATED) + return 0; + + actor = rpcsvc_program_actor(req); + if (!actor) + return 0; + + return (actor->op_type == DRC_NON_IDEMPOTENT && drc->type != DRC_TYPE_NONE); +} + +/** + * rpcsvc_drc_client_ref - ref the drc client + * + * @param client - the drc client to ref + * @return client + */ +static drc_client_t * +rpcsvc_drc_client_ref(drc_client_t *client) +{ + GF_ASSERT(client); + GF_ATOMIC_INC(client->ref); + return client; +} + +/** + * rpcsvc_drc_client_unref - unref the drc client, and destroy + * the client on last unref + * + * @param drc - the main drc structure + * @param client - the drc client to unref + * @return NULL if it is the last unref, client otherwise + */ +static drc_client_t * +rpcsvc_drc_client_unref(rpcsvc_drc_globals_t *drc, drc_client_t *client) +{ + uint32_t refcount; + + GF_ASSERT(drc); + + refcount = GF_ATOMIC_DEC(client->ref); + if (!refcount) { + drc->client_count--; + rpcsvc_remove_drc_client(client); + client = NULL; + } + + return client; +} + +/** + * rpcsvc_drc_lookup - lookup a request to see if it is already cached + * + * @param req - incoming request + * @return cached reply of req if found, NULL otherwise + */ +drc_cached_op_t * +rpcsvc_drc_lookup(rpcsvc_request_t *req) +{ + drc_client_t *client = NULL; + drc_cached_op_t *reply = NULL; + drc_cached_op_t new = { + .xid = req->xid, + .prognum = req->prognum, + .progversion = req->progver, + .procnum = req->procnum, + }; + + GF_ASSERT(req); + + if (!req->trans->drc_client) { + client = rpcsvc_get_drc_client(req->svc->drc, + &req->trans->peerinfo.sockaddr); + if (!client) + goto out; + + req->trans->drc_client = rpcsvc_drc_client_ref(client); + } + + client = req->trans->drc_client; + + if (client->op_count == 0) + goto out; + + reply = rb_find(client->rbtree, &new); + +out: + return reply; +} + +/** + * rpcsvc_send_cached_reply - send the cached reply for the incoming request + * + * @param req - incoming request (which is a duplicate in this case) + * @param reply - the cached reply for req + * @return 0 on successful reply submission, -1 or other non-zero value + * otherwise + */ +int +rpcsvc_send_cached_reply(rpcsvc_request_t *req, drc_cached_op_t *reply) +{ + int ret = 0; + + GF_ASSERT(req); + GF_ASSERT(reply); + + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "sending cached reply: xid: %d, " + "client: %s", + req->xid, req->trans->peerinfo.identifier); + + rpcsvc_drc_client_ref(reply->client); + ret = rpcsvc_transport_submit( + req->trans, reply->msg.rpchdr, reply->msg.rpchdrcount, + reply->msg.proghdr, reply->msg.proghdrcount, reply->msg.progpayload, + reply->msg.progpayloadcount, reply->msg.iobref, req->trans_private); + rpcsvc_drc_client_unref(req->svc->drc, reply->client); + + return ret; +} + +/** + * rpcsvc_cache_reply - cache the reply for the processed request 'req' + * + * @param req - processed request + * @param iobref - iobref structure of the reply + * @param rpchdr - rpc header of the reply + * @param rpchdrcount - size of rpchdr + * @param proghdr - program header of the reply + * @param proghdrcount - size of proghdr + * @param payload - payload of the reply if any + * @param payloadcount - size of payload + * @return 0 on success, -1 on failure + */ +int +rpcsvc_cache_reply(rpcsvc_request_t *req, struct iobref *iobref, + struct iovec *rpchdr, int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *payload, int payloadcount) +{ + int ret = -1; + drc_cached_op_t *reply = NULL; + + GF_ASSERT(req); + GF_ASSERT(req->reply); + + reply = req->reply; + + reply->state = DRC_OP_CACHED; + + reply->msg.iobref = iobref_ref(iobref); + + reply->msg.rpchdrcount = rpchdrcount; + reply->msg.rpchdr = iov_dup(rpchdr, rpchdrcount); + + reply->msg.proghdrcount = proghdrcount; + reply->msg.proghdr = iov_dup(proghdr, proghdrcount); + + reply->msg.progpayloadcount = payloadcount; + if (payloadcount) + reply->msg.progpayload = iov_dup(payload, payloadcount); + + // rpcsvc_drc_client_unref (req->svc->drc, req->trans->drc_client); + // rpcsvc_drc_op_unref (req->svc->drc, reply); + ret = 0; + + return ret; +} + +/** + * rpcsvc_vacate_drc_entries - free up some percentage of drc cache + * based on the lru factor + * + * @param drc - the main drc structure + * @return void + */ +static void +rpcsvc_vacate_drc_entries(rpcsvc_drc_globals_t *drc) +{ + uint32_t i = 0; + uint32_t n = 0; + drc_cached_op_t *reply = NULL; + drc_cached_op_t *tmp = NULL; + drc_client_t *client = NULL; + + GF_ASSERT(drc); + + n = drc->global_cache_size / drc->lru_factor; + + list_for_each_entry_safe_reverse(reply, tmp, &drc->cache_head, global_list) + { + /* Don't delete ops that are in transit */ + if (reply->state == DRC_OP_IN_TRANSIT) + continue; + + client = reply->client; + + rb_delete(client->rbtree, reply); + + rpcsvc_drc_op_destroy(drc, reply); + rpcsvc_drc_client_unref(drc, client); + i++; + if (i >= n) + break; + } +} + +/** + * rpcsvc_add_op_to_cache - insert the cached op into the client rbtree and drc + * list + * + * @param drc - the main drc structure + * @param reply - the op to be inserted + * @return 0 on success, -1 on failure + */ +static int +rpcsvc_add_op_to_cache(rpcsvc_drc_globals_t *drc, drc_cached_op_t *reply) +{ + drc_client_t *client = NULL; + drc_cached_op_t **tmp_reply = NULL; + + GF_ASSERT(drc); + GF_ASSERT(reply); + + client = reply->client; + + /* cache is full, free up some space */ + if (drc->op_count >= drc->global_cache_size) + rpcsvc_vacate_drc_entries(drc); + + tmp_reply = (drc_cached_op_t **)rb_probe(client->rbtree, reply); + if (!tmp_reply) { + /* mem alloc failed */ + return -1; + } else if (*tmp_reply != reply) { + /* should never happen */ + gf_log(GF_RPCSVC, GF_LOG_ERROR, "DRC failed to detect duplicates"); + return -1; + } + + client->op_count++; + list_add(&reply->global_list, &drc->cache_head); + drc->op_count++; + + return 0; +} + +/** + * rpcsvc_cache_request - cache the in-transition incoming request + * + * @param req - incoming request + * @return 0 on success, -1 on failure + */ +int +rpcsvc_cache_request(rpcsvc_request_t *req) +{ + int ret = -1; + drc_client_t *client = NULL; + drc_cached_op_t *reply = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT(req); + + drc = req->svc->drc; + + client = req->trans->drc_client; + if (!client) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "drc client is NULL"); + goto out; + } + + reply = mem_get0(drc->mempool); + if (!reply) + goto out; + + reply->client = rpcsvc_drc_client_ref(client); + reply->xid = req->xid; + reply->prognum = req->prognum; + reply->progversion = req->progver; + reply->procnum = req->procnum; + reply->state = DRC_OP_IN_TRANSIT; + req->reply = reply; + INIT_LIST_HEAD(&reply->global_list); + + ret = rpcsvc_add_op_to_cache(drc, reply); + if (ret) { + req->reply = NULL; + rpcsvc_drc_op_destroy(drc, reply); + rpcsvc_drc_client_unref(drc, client); + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "Failed to add op to drc cache"); + } + +out: + return ret; +} + +/** + * + * rpcsvc_drc_priv - function which dumps the drc state + * + * @param drc - the main drc structure + * @return 0 on success, -1 on failure + */ +int32_t +rpcsvc_drc_priv(rpcsvc_drc_globals_t *drc) +{ + int i = 0; + char key[GF_DUMP_MAX_BUF_LEN] = {0}; + drc_client_t *client = NULL; + char ip[INET6_ADDRSTRLEN] = {0}; + + if (!drc || drc->status == DRC_UNINITIATED) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "DRC is " + "uninitialized, not dumping its state"); + return 0; + } + + gf_proc_dump_add_section("rpc.drc"); + + if (TRY_LOCK(&drc->lock)) + return -1; + + gf_proc_dump_build_key(key, "drc", "type"); + gf_proc_dump_write(key, "%d", drc->type); + + gf_proc_dump_build_key(key, "drc", "client_count"); + gf_proc_dump_write(key, "%d", drc->client_count); + + gf_proc_dump_build_key(key, "drc", "current_cache_size"); + gf_proc_dump_write(key, "%d", drc->op_count); + + gf_proc_dump_build_key(key, "drc", "max_cache_size"); + gf_proc_dump_write(key, "%d", drc->global_cache_size); + + gf_proc_dump_build_key(key, "drc", "lru_factor"); + gf_proc_dump_write(key, "%d", drc->lru_factor); + + gf_proc_dump_build_key(key, "drc", "duplicate_request_count"); + gf_proc_dump_write(key, "%" PRIu64, drc->cache_hits); + + gf_proc_dump_build_key(key, "drc", "in_transit_duplicate_requests"); + gf_proc_dump_write(key, "%" PRIu64, drc->intransit_hits); + + list_for_each_entry(client, &drc->clients_head, client_list) + { + gf_proc_dump_build_key(key, "client", "%d.ip-address", i); + memset(ip, 0, INET6_ADDRSTRLEN); + switch (client->sock_union.storage.ss_family) { + case AF_INET: + gf_proc_dump_write( + key, "%s", + inet_ntop(AF_INET, &client->sock_union.sin.sin_addr.s_addr, + ip, INET_ADDRSTRLEN)); + break; + case AF_INET6: + gf_proc_dump_write( + key, "%s", + inet_ntop(AF_INET6, &client->sock_union.sin6.sin6_addr, ip, + INET6_ADDRSTRLEN)); + break; + default: + gf_proc_dump_write(key, "%s", "N/A"); + } + + gf_proc_dump_build_key(key, "client", "%d.ref_count", i); + gf_proc_dump_write(key, "%" PRIu32, GF_ATOMIC_GET(client->ref)); + gf_proc_dump_build_key(key, "client", "%d.op_count", i); + gf_proc_dump_write(key, "%d", client->op_count); + i++; + } + + UNLOCK(&drc->lock); + return 0; +} + +/** + * rpcsvc_drc_notify - function which is notified of RPC transport events + * + * @param svc - pointer to rpcsvc_t structure of the rpc + * @param xl - pointer to the xlator + * @param event - the event which triggered this notify + * @param data - the transport structure + * @return 0 on success, -1 on failure + */ +int +rpcsvc_drc_notify(rpcsvc_t *svc, void *xl, rpcsvc_event_t event, void *data) +{ + int ret = -1; + rpc_transport_t *trans = NULL; + drc_client_t *client = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT(svc); + GF_ASSERT(svc->drc); + GF_ASSERT(data); + + drc = svc->drc; + + if (drc->status == DRC_UNINITIATED || drc->type == DRC_TYPE_NONE) + return 0; + + LOCK(&drc->lock); + { + trans = (rpc_transport_t *)data; + client = rpcsvc_get_drc_client(drc, &trans->peerinfo.sockaddr); + if (!client) + goto unlock; + + switch (event) { + case RPCSVC_EVENT_ACCEPT: + trans->drc_client = rpcsvc_drc_client_ref(client); + ret = 0; + break; + + case RPCSVC_EVENT_DISCONNECT: + ret = 0; + if (list_empty(&drc->clients_head)) + break; + /* should be the last unref */ + trans->drc_client = NULL; + rpcsvc_drc_client_unref(drc, client); + break; + + default: + break; + } + } +unlock: + UNLOCK(&drc->lock); + return ret; +} + +/** + * rpcsvc_drc_init - Initialize the duplicate request cache service + * + * @param svc - pointer to rpcsvc_t structure of the rpc + * @param options - the options dictionary which configures drc + * @return 0 on success, non-zero integer on failure + */ +int +rpcsvc_drc_init(rpcsvc_t *svc, dict_t *options) +{ + int ret = 0; + uint32_t drc_type = 0; + uint32_t drc_size = 0; + uint32_t drc_factor = 0; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT(svc); + GF_ASSERT(options); + + /* Toggle DRC on/off, when more drc types(persistent/cluster) + * are added, we shouldn't treat this as boolean. */ + ret = dict_get_str_boolean(options, "nfs.drc", _gf_false); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_INFO, "drc user options need second look"); + ret = _gf_false; + } + + gf_log(GF_RPCSVC, GF_LOG_INFO, "DRC is turned %s", (ret ? "ON" : "OFF")); + + /*DRC off, nothing to do */ + if (ret == _gf_false) + return (0); + + drc = GF_CALLOC(1, sizeof(rpcsvc_drc_globals_t), + gf_common_mt_drc_globals_t); + if (!drc) + return (-1); + + LOCK_INIT(&drc->lock); + svc->drc = drc; + + /* Specify type of DRC to be used */ + ret = dict_get_uint32(options, "nfs.drc-type", &drc_type); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "drc type not set. Continuing with default"); + drc_type = DRC_DEFAULT_TYPE; + } + + /* Set the global cache size (no. of ops to cache) */ + ret = dict_get_uint32(options, "nfs.drc-size", &drc_size); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "drc size not set. Continuing with default size"); + drc_size = DRC_DEFAULT_CACHE_SIZE; + } + + LOCK(&drc->lock); + + drc->type = drc_type; + drc->global_cache_size = drc_size; + + /* Mempool for cached ops */ + drc->mempool = mem_pool_new(drc_cached_op_t, drc->global_cache_size); + if (!drc->mempool) { + UNLOCK(&drc->lock); + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to get mempool for DRC, drc-size: %d", drc_size); + ret = -1; + goto post_unlock; + } + + /* What percent of cache to be evicted whenever it fills up */ + ret = dict_get_uint32(options, "nfs.drc-lru-factor", &drc_factor); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "drc lru factor not set. Continuing with policy default"); + drc_factor = DRC_DEFAULT_LRU_FACTOR; + } + + drc->lru_factor = (drc_lru_factor_t)drc_factor; + + INIT_LIST_HEAD(&drc->clients_head); + INIT_LIST_HEAD(&drc->cache_head); + + ret = rpcsvc_register_notify(svc, rpcsvc_drc_notify, THIS); + if (ret) { + UNLOCK(&drc->lock); + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "registration of drc_notify function failed"); + goto post_unlock; + } + + drc->status = DRC_INITIATED; + UNLOCK(&drc->lock); + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "drc init successful"); +post_unlock: + if (ret == -1) { + if (drc->mempool) { + mem_pool_destroy(drc->mempool); + drc->mempool = NULL; + } + GF_FREE(drc); + svc->drc = NULL; + } + return ret; +} + +int +rpcsvc_drc_deinit(rpcsvc_t *svc) +{ + rpcsvc_drc_globals_t *drc = NULL; + + if (!svc) + return (-1); + + drc = svc->drc; + if (!drc) + return (0); + + LOCK(&drc->lock); + (void)rpcsvc_unregister_notify(svc, rpcsvc_drc_notify, THIS); + if (drc->mempool) { + mem_pool_destroy(drc->mempool); + drc->mempool = NULL; + } + UNLOCK(&drc->lock); + + GF_FREE(drc); + svc->drc = NULL; + + return (0); +} + +int +rpcsvc_drc_reconfigure(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + gf_boolean_t enable_drc = _gf_false; + rpcsvc_drc_globals_t *drc = NULL; + uint32_t drc_size = 0; + + /* Input sanitization */ + if ((!svc) || (!options)) + return (-1); + + /* If DRC was not enabled before, Let rpcsvc_drc_init() to + * take care of DRC initialization part. + */ + drc = svc->drc; + if (!drc) { + return rpcsvc_drc_init(svc, options); + } + + /* DRC was already enabled before. Going to be reconfigured. Check + * if reconfigured options contain "nfs.drc" and "nfs.drc-size". + * + * NB: If DRC is "OFF", "drc-size" has no role to play. + * So, "drc-size" gets evaluated IFF DRC is "ON". + * + * If DRC is reconfigured, + * case 1: DRC is "ON" + * sub-case 1: drc-size remains same + * ACTION: Nothing to do. + * sub-case 2: drc-size just changed + * ACTION: rpcsvc_drc_deinit() followed by + * rpcsvc_drc_init(). + * + * case 2: DRC is "OFF" + * ACTION: rpcsvc_drc_deinit() + */ + ret = dict_get_str_boolean(options, "nfs.drc", _gf_false); + if (ret < 0) + ret = _gf_false; + + enable_drc = ret; + gf_log(GF_RPCSVC, GF_LOG_INFO, "DRC is turned %s", (ret ? "ON" : "OFF")); + + /* case 1: DRC is "ON"*/ + if (enable_drc) { + /* Fetch drc-size if reconfigured */ + if (dict_get_uint32(options, "nfs.drc-size", &drc_size)) + drc_size = DRC_DEFAULT_CACHE_SIZE; + + /* case 1: sub-case 1*/ + if (drc->global_cache_size == drc_size) + return (0); + + /* case 1: sub-case 2*/ + (void)rpcsvc_drc_deinit(svc); + return rpcsvc_drc_init(svc, options); + } + + /* case 2: DRC is "OFF" */ + return rpcsvc_drc_deinit(svc); +} diff --git a/rpc/rpc-lib/src/rpc-drc.h b/rpc/rpc-lib/src/rpc-drc.h new file mode 100644 index 00000000000..ce66430809b --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.h @@ -0,0 +1,97 @@ +/* + Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef RPC_DRC_H +#define RPC_DRC_H + +#include "rpcsvc-common.h" +#include "rpcsvc.h" +#include <glusterfs/locking.h> +#include <glusterfs/dict.h> +#include "rb.h" + +/* per-client cache structure */ +struct drc_client { + union gf_sock_union sock_union; + /* pointers to the cache */ + struct rb_table *rbtree; + /* no. of ops currently cached */ + uint32_t op_count; + gf_atomic_uint32_t ref; + struct list_head client_list; +}; + +struct drc_cached_op { + drc_op_state_t state; + int prognum; + int progversion; + int procnum; + rpc_transport_msg_t msg; + drc_client_t *client; + struct list_head client_list; + struct list_head global_list; + int32_t ref; + uint32_t xid; +}; + +/* global drc definitions */ +enum drc_status { DRC_UNINITIATED, DRC_INITIATED }; +typedef enum drc_status drc_status_t; + +struct drc_globals { + /* allocator must be the first member since + * it is used so in gf_libavl_allocator + */ + struct libavl_allocator allocator; + /* configurable size parameter */ + gf_lock_t lock; + uint64_t cache_hits; + uint64_t intransit_hits; + struct mem_pool *mempool; + struct list_head cache_head; + struct list_head clients_head; + uint32_t op_count; + uint32_t client_count; + uint32_t global_cache_size; + drc_type_t type; + drc_lru_factor_t lru_factor; + drc_status_t status; +}; + +int +rpcsvc_need_drc(rpcsvc_request_t *req); + +drc_cached_op_t * +rpcsvc_drc_lookup(rpcsvc_request_t *req); + +int +rpcsvc_send_cached_reply(rpcsvc_request_t *req, drc_cached_op_t *reply); + +int +rpcsvc_cache_reply(rpcsvc_request_t *req, struct iobref *iobref, + struct iovec *rpchdr, int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *payload, int payloadcount); + +int +rpcsvc_cache_request(rpcsvc_request_t *req); + +int32_t +rpcsvc_drc_priv(rpcsvc_drc_globals_t *drc); + +int +rpcsvc_drc_init(rpcsvc_t *svc, dict_t *options); + +int +rpcsvc_drc_deinit(rpcsvc_t *svc); + +int +rpcsvc_drc_reconfigure(rpcsvc_t *svc, dict_t *options); + +#endif /* RPC_DRC_H */ diff --git a/rpc/rpc-lib/src/rpc-lib-messages.h b/rpc/rpc-lib/src/rpc-lib-messages.h new file mode 100644 index 00000000000..2c0b820dbf9 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-lib-messages.h @@ -0,0 +1,34 @@ +/* + Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _RPC_LIB_MESSAGES_H_ +#define _RPC_LIB_MESSAGES_H_ + +#include <glusterfs/glfs-message-id.h> + +/* To add new message IDs, append new identifiers at the end of the list. + * + * Never remove a message ID. If it's not used anymore, you can rename it or + * leave it as it is, but not delete it. This is to prevent reutilization of + * IDs by other messages. + * + * The component name must match one of the entries defined in + * glfs-message-id.h. + */ + +GLFS_MSGID(RPC_LIB, TRANS_MSG_ADDR_FAMILY_NOT_SPECIFIED, + TRANS_MSG_UNKNOWN_ADDR_FAMILY, TRANS_MSG_REMOTE_HOST_ERROR, + TRANS_MSG_DNS_RESOL_FAILED, TRANS_MSG_LISTEN_PATH_ERROR, + TRANS_MSG_CONNECT_PATH_ERROR, TRANS_MSG_GET_ADDR_INFO_FAILED, + TRANS_MSG_PORT_BIND_FAILED, TRANS_MSG_INET_ERROR, + TRANS_MSG_GET_NAME_INFO_FAILED, TRANS_MSG_TRANSPORT_ERROR, + TRANS_MSG_TIMEOUT_EXCEEDED, TRANS_MSG_SOCKET_BIND_ERROR); + +#endif /* !_RPC_LIB_MESSAGES_H_ */ diff --git a/rpc/rpc-lib/src/rpc-transport.c b/rpc/rpc-lib/src/rpc-transport.c new file mode 100644 index 00000000000..a6e201a9b36 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-transport.c @@ -0,0 +1,672 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include <dlfcn.h> +#include <stdlib.h> +#include <stdio.h> +#include <sys/poll.h> +#include <stdint.h> + +#include "rpc-transport.h" + +#ifndef GF_OPTION_LIST_EMPTY +#define GF_OPTION_LIST_EMPTY(_opt) (_opt->value[0] == NULL) +#endif + +int32_t +rpc_transport_count(const char *transport_type) +{ + char *transport_dup = NULL; + char *saveptr = NULL; + char *ptr = NULL; + int count = 0; + + if (transport_type == NULL) + return -1; + + transport_dup = gf_strdup(transport_type); + if (transport_dup == NULL) { + return -1; + } + + ptr = strtok_r(transport_dup, ",", &saveptr); + while (ptr != NULL) { + count++; + ptr = strtok_r(NULL, ",", &saveptr); + } + + GF_FREE(transport_dup); + return count; +} + +int +rpc_transport_get_myaddr(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, size_t salen) +{ + int32_t ret = -1; + GF_VALIDATE_OR_GOTO("rpc", this, out); + + ret = this->ops->get_myaddr(this, peeraddr, addrlen, sa, salen); + +out: + return ret; +} + +int32_t +rpc_transport_get_peername(rpc_transport_t *this, char *hostname, int hostlen) +{ + int32_t ret = -1; + GF_VALIDATE_OR_GOTO("rpc", this, out); + + ret = this->ops->get_peername(this, hostname, hostlen); +out: + return ret; +} + +int +rpc_transport_throttle(rpc_transport_t *this, gf_boolean_t onoff) +{ + if (!this->ops->throttle) + return -ENOSYS; + + return this->ops->throttle(this, onoff); +} + +int32_t +rpc_transport_get_peeraddr(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, size_t salen) +{ + int32_t ret = -1; + GF_VALIDATE_OR_GOTO("rpc", this, out); + + ret = this->ops->get_peeraddr(this, peeraddr, addrlen, sa, salen); +out: + return ret; +} + +void +rpc_transport_pollin_destroy(rpc_transport_pollin_t *pollin) +{ + GF_VALIDATE_OR_GOTO("rpc", pollin, out); + + if (pollin->iobref) { + iobref_unref(pollin->iobref); + } + + if (pollin->private) { + /* */ + GF_FREE(pollin->private); + } + + GF_FREE(pollin); +out: + return; +} + +rpc_transport_pollin_t * +rpc_transport_pollin_alloc(rpc_transport_t *this, struct iovec *vector, + int count, struct iobuf *hdr_iobuf, + struct iobref *iobref, void *private) +{ + rpc_transport_pollin_t *msg = NULL; + msg = GF_CALLOC(1, sizeof(*msg), gf_common_mt_rpc_trans_pollin_t); + if (!msg) { + goto out; + } + + msg->trans = this; + + if (count > 1) { + msg->vectored = 1; + } + + memcpy(msg->vector, vector, count * sizeof(*vector)); + msg->count = count; + msg->iobref = iobref_ref(iobref); + msg->private = private; + if (hdr_iobuf) + iobref_add(iobref, hdr_iobuf); + +out: + return msg; +} + +void +rpc_transport_cleanup(rpc_transport_t *trans) +{ + if (!trans) + return; + + if (trans->fini) + trans->fini(trans); + + if (trans->options) { + dict_unref(trans->options); + trans->options = NULL; + } + + GF_FREE(trans->name); + + if (trans->xl) + pthread_mutex_destroy(&trans->lock); + + if (trans->dl_handle) + dlclose(trans->dl_handle); + + GF_FREE(trans); +} + +rpc_transport_t * +rpc_transport_load(glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) +{ + struct rpc_transport *trans = NULL, *return_trans = NULL; + char *name = NULL; + void *handle = NULL; + char *type = NULL; + static char str[] = "ERROR"; + int32_t ret = -1; + int is_tcp = 0, is_unix = 0, is_ibsdp = 0; + volume_opt_list_t *vol_opt = NULL; + gf_boolean_t bind_insecure = _gf_false; + xlator_t *this = NULL; + gf_boolean_t success = _gf_false; + + GF_VALIDATE_OR_GOTO("rpc-transport", options, fail); + GF_VALIDATE_OR_GOTO("rpc-transport", ctx, fail); + GF_VALIDATE_OR_GOTO("rpc-transport", trans_name, fail); + + trans = GF_CALLOC(1, sizeof(struct rpc_transport), + gf_common_mt_rpc_trans_t); + if (!trans) + goto fail; + + trans->name = gf_strdup(trans_name); + if (!trans->name) + goto fail; + + trans->ctx = ctx; + type = str; + + /* Backward compatibility */ + ret = dict_get_str_sizen(options, "transport-type", &type); + if (ret < 0) { + ret = dict_set_str_sizen(options, "transport-type", "socket"); + if (ret < 0) + gf_log("dict", GF_LOG_DEBUG, "setting transport-type failed"); + else + gf_log("rpc-transport", GF_LOG_DEBUG, + "missing 'option transport-type'. defaulting to " + "\"socket\""); + } else { + { + /* Backward compatibility to handle * /client, + * * /server. + */ + char *tmp = strchr(type, '/'); + if (tmp) + *tmp = '\0'; + } + + is_tcp = strcmp(type, "tcp"); + is_unix = strcmp(type, "unix"); + is_ibsdp = strcmp(type, "ib-sdp"); + if ((is_tcp == 0) || (is_unix == 0) || (is_ibsdp == 0)) { + if (is_unix == 0) + ret = dict_set_str_sizen(options, "transport.address-family", + "unix"); + if (is_ibsdp == 0) + ret = dict_set_str_sizen(options, "transport.address-family", + "inet-sdp"); + + if (ret < 0) + gf_log("dict", GF_LOG_DEBUG, "setting address-family failed"); + + ret = dict_set_str_sizen(options, "transport-type", "socket"); + if (ret < 0) + gf_log("dict", GF_LOG_DEBUG, "setting transport-type failed"); + } + } + + /* client-bind-insecure is for clients protocol, and + * bind-insecure for glusterd. Both mutually exclusive + */ + ret = dict_get_str_sizen(options, "client-bind-insecure", &type); + if (ret) + ret = dict_get_str_sizen(options, "bind-insecure", &type); + if (ret == 0) { + ret = gf_string2boolean(type, &bind_insecure); + if (ret < 0) { + gf_log("rcp-transport", GF_LOG_WARNING, + "bind-insecure option %s is not a" + " valid bool option", + type); + goto fail; + } + if (_gf_true == bind_insecure) + trans->bind_insecure = 1; + else + trans->bind_insecure = 0; + } else { + /* By default allow bind insecure */ + trans->bind_insecure = 1; + } + + ret = dict_get_str_sizen(options, "transport-type", &type); + if (ret < 0) { + gf_log("rpc-transport", GF_LOG_ERROR, + "'option transport-type <xx>' missing in volume '%s'", + trans_name); + goto fail; + } + + ret = gf_asprintf(&name, "%s/%s.so", RPC_TRANSPORTDIR, type); + if (-1 == ret) { + goto fail; + } + + if (dict_get_sizen(options, "notify-poller-death")) { + trans->notify_poller_death = 1; + } + + gf_log("rpc-transport", GF_LOG_DEBUG, "attempt to load file %s", name); + + handle = dlopen(name, RTLD_NOW); + if (handle == NULL) { + gf_log("rpc-transport", GF_LOG_ERROR, "%s", dlerror()); + gf_log("rpc-transport", GF_LOG_WARNING, + "volume '%s': transport-type '%s' is not valid or " + "not found on this machine", + trans_name, type); + goto fail; + } + + trans->dl_handle = handle; + + trans->ops = dlsym(handle, "tops"); + if (trans->ops == NULL) { + gf_log("rpc-transport", GF_LOG_ERROR, "dlsym (rpc_transport_ops) on %s", + dlerror()); + goto fail; + } + + *VOID(&(trans->init)) = dlsym(handle, "init"); + if (trans->init == NULL) { + gf_log("rpc-transport", GF_LOG_ERROR, + "dlsym (gf_rpc_transport_init) on %s", dlerror()); + goto fail; + } + + *VOID(&(trans->fini)) = dlsym(handle, "fini"); + if (trans->fini == NULL) { + gf_log("rpc-transport", GF_LOG_ERROR, + "dlsym (gf_rpc_transport_fini) on %s", dlerror()); + goto fail; + } + + *VOID(&(trans->reconfigure)) = dlsym(handle, "reconfigure"); + if (trans->reconfigure == NULL) { + gf_log("rpc-transport", GF_LOG_DEBUG, + "dlsym (gf_rpc_transport_reconfigure) on %s", dlerror()); + } + + vol_opt = GF_CALLOC(1, sizeof(volume_opt_list_t), + gf_common_mt_volume_opt_list_t); + if (!vol_opt) { + goto fail; + } + + this = THIS; + vol_opt->given_opt = dlsym(handle, "options"); + if (vol_opt->given_opt == NULL) { + gf_log("rpc-transport", GF_LOG_DEBUG, + "volume option validation not specified"); + } else { + INIT_LIST_HEAD(&vol_opt->list); + list_add_tail(&vol_opt->list, &(this->volume_options)); + if (xlator_options_validate_list(this, options, vol_opt, NULL)) { + gf_log("rpc-transport", GF_LOG_ERROR, + "volume option validation failed"); + goto fail; + } + } + + trans->options = dict_ref(options); + + pthread_mutex_init(&trans->lock, NULL); + trans->xl = this; + + ret = trans->init(trans); + if (ret != 0) { + gf_log("rpc-transport", GF_LOG_WARNING, "'%s' initialization failed", + type); + goto fail; + } + + INIT_LIST_HEAD(&trans->list); + GF_ATOMIC_INIT(trans->disconnect_progress, 0); + + return_trans = trans; + + GF_FREE(name); + + success = _gf_true; + +fail: + if (!success) { + rpc_transport_cleanup(trans); + GF_FREE(name); + + return_trans = NULL; + } + + if (vol_opt) { + if (!list_empty(&vol_opt->list)) { + list_del_init(&vol_opt->list); + } + GF_FREE(vol_opt); + } + + return return_trans; +} + +int32_t +rpc_transport_submit_request(rpc_transport_t *this, rpc_transport_req_t *req) +{ + int32_t ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + GF_VALIDATE_OR_GOTO("rpc_transport", this->ops, fail); + + ret = this->ops->submit_request(this, req); +fail: + return ret; +} + +int32_t +rpc_transport_submit_reply(rpc_transport_t *this, rpc_transport_reply_t *reply) +{ + int32_t ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + GF_VALIDATE_OR_GOTO("rpc_transport", this->ops, fail); + + ret = this->ops->submit_reply(this, reply); +fail: + return ret; +} + +int32_t +rpc_transport_connect(rpc_transport_t *this, int port) +{ + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + + ret = this->ops->connect(this, port); +fail: + return ret; +} + +int32_t +rpc_transport_listen(rpc_transport_t *this) +{ + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + + ret = this->ops->listen(this); +fail: + return ret; +} + +int32_t +rpc_transport_disconnect(rpc_transport_t *this, gf_boolean_t wait) +{ + int32_t ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + + ret = this->ops->disconnect(this, wait); + +fail: + return ret; +} + +static void +rpc_transport_destroy(rpc_transport_t *this) +{ + struct dnscache6 *cache = NULL; + + if (this->clnt_options) + dict_unref(this->clnt_options); + if (this->options) + dict_unref(this->options); + if (this->fini) + this->fini(this); + + pthread_mutex_destroy(&this->lock); + + GF_FREE(this->name); + + if (this->dl_handle) + dlclose(this->dl_handle); + + if (this->ssl_name) { + GF_FREE(this->ssl_name); + } + + if (this->dnscache) { + cache = this->dnscache; + if (cache->first) + freeaddrinfo(cache->first); + GF_FREE(this->dnscache); + } + + GF_FREE(this); +} + +rpc_transport_t * +rpc_transport_ref(rpc_transport_t *this) +{ + rpc_transport_t *return_this = NULL; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + + GF_ATOMIC_INC(this->refcount); + + return_this = this; +fail: + return return_this; +} + +int32_t +rpc_transport_unref(rpc_transport_t *this) +{ + int32_t refcount = 0; + int32_t ret = -1; + + GF_VALIDATE_OR_GOTO("rpc_transport", this, fail); + + refcount = GF_ATOMIC_DEC(this->refcount); + + if (refcount == 0) { + if (this->mydata) + this->notify(this, this->mydata, RPC_TRANSPORT_CLEANUP, NULL); + this->mydata = NULL; + this->notify = NULL; + rpc_transport_destroy(this); + } + + ret = 0; +fail: + return ret; +} + +int32_t +rpc_transport_notify(rpc_transport_t *this, rpc_transport_event_t event, + void *data, ...) +{ + int32_t ret = -1; + GF_VALIDATE_OR_GOTO("rpc", this, out); + + if (this->notify != NULL) { + ret = this->notify(this, this->mydata, event, data); + } else { + ret = 0; + } +out: + return ret; +} + +int +rpc_transport_register_notify(rpc_transport_t *trans, + rpc_transport_notify_t notify, void *mydata) +{ + int32_t ret = -1; + GF_VALIDATE_OR_GOTO("rpc", trans, out); + + trans->notify = notify; + trans->mydata = mydata; + + ret = 0; +out: + return ret; +} + +// give negative values to skip setting that value +// this function asserts if both the values are negative. +// why call it if you don't set it. +int +rpc_transport_keepalive_options_set(dict_t *options, int32_t interval, + int32_t time, int32_t timeout) +{ + int ret = -1; + + GF_ASSERT(options); + GF_ASSERT((interval > 0) || (time > 0)); + + ret = dict_set_int32_sizen(options, "transport.socket.keepalive-interval", + interval); + if (ret) + goto out; + + ret = dict_set_int32_sizen(options, "transport.socket.keepalive-time", + time); + if (ret) + goto out; + + ret = dict_set_int32_sizen(options, "transport.tcp-user-timeout", timeout); + if (ret) + goto out; +out: + return ret; +} + +int +rpc_transport_unix_options_build(dict_t *dict, char *filepath, + int frame_timeout) +{ + char *fpath = NULL; + int ret = -1; + + GF_ASSERT(filepath); + GF_VALIDATE_OR_GOTO("rpc-transport", dict, out); + + fpath = gf_strdup(filepath); + if (!fpath) { + ret = -1; + goto out; + } + + ret = dict_set_dynstr_sizen(dict, "transport.socket.connect-path", fpath); + if (ret) { + GF_FREE(fpath); + goto out; + } + + ret = dict_set_str_sizen(dict, "transport.address-family", "unix"); + if (ret) + goto out; + + ret = dict_set_str_sizen(dict, "transport.socket.nodelay", "off"); + if (ret) + goto out; + + ret = dict_set_str_sizen(dict, "transport-type", "socket"); + if (ret) + goto out; + + ret = dict_set_str_sizen(dict, "transport.socket.keepalive", "off"); + if (ret) + goto out; + + if (frame_timeout > 0) { + ret = dict_set_int32_sizen(dict, "frame-timeout", frame_timeout); + if (ret) + goto out; + } +out: + return ret; +} + +int +rpc_transport_inet_options_build(dict_t *dict, const char *hostname, int port, + char *af) +{ + char *host = NULL; + int ret = -1; +#ifdef IPV6_DEFAULT + static char *addr_family = "inet6"; +#else + static char *addr_family = "inet"; +#endif + + GF_ASSERT(hostname); + GF_ASSERT(port >= 1024); + GF_VALIDATE_OR_GOTO("rpc-transport", dict, out); + + host = gf_strdup((char *)hostname); + if (!host) { + ret = -1; + goto out; + } + + ret = dict_set_dynstr_sizen(dict, "remote-host", host); + if (ret) { + gf_log(THIS->name, GF_LOG_WARNING, "failed to set remote-host with %s", + host); + GF_FREE(host); + goto out; + } + + ret = dict_set_int32_sizen(dict, "remote-port", port); + if (ret) { + gf_log(THIS->name, GF_LOG_WARNING, "failed to set remote-port with %d", + port); + goto out; + } + + ret = dict_set_str_sizen(dict, "address-family", + (af != NULL ? af : addr_family)); + if (ret) { + gf_log(THIS->name, GF_LOG_WARNING, "failed to set address-family to %s", + addr_family); + goto out; + } + + ret = dict_set_str_sizen(dict, "transport-type", "socket"); + if (ret) { + gf_log(THIS->name, GF_LOG_WARNING, + "failed to set trans-type with socket"); + goto out; + } +out: + return ret; +} diff --git a/rpc/rpc-lib/src/rpc-transport.h b/rpc/rpc-lib/src/rpc-transport.h new file mode 100644 index 00000000000..c499f0bb955 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-transport.h @@ -0,0 +1,312 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef __RPC_TRANSPORT_H__ +#define __RPC_TRANSPORT_H__ + +#include <inttypes.h> +#ifdef GF_SOLARIS_HOST_OS +#include <rpc/auth.h> +#else +#include <rpc/rpc.h> +#endif + +#include <rpc/rpc_msg.h> + +#ifndef MAX_IOVEC +#define MAX_IOVEC 16 +#endif + +#ifndef AI_ADDRCONFIG +#define AI_ADDRCONFIG 0 +#endif /* AI_ADDRCONFIG */ + +/* Given the 4-byte fragment header, returns non-zero if this fragment + * is the last fragment for the RPC record being assembled. + * RPC Record marking standard defines a 32 bit value as the fragment + * header with the MSB signifying whether the fragment is the last + * fragment for the record being assembled. + */ +#define RPC_LASTFRAG(fraghdr) ((uint32_t)(fraghdr & 0x80000000U)) + +/* Given the 4-byte fragment header, extracts the bits that contain + * the fragment size. + */ +#define RPC_FRAGSIZE(fraghdr) ((uint32_t)(fraghdr & 0x7fffffffU)) + +#define RPC_FRAGHDR_SIZE 4 +#define RPC_MSGTYPE_SIZE 8 + +/* size of the msg from the start of call-body till and including credlen */ +#define RPC_CALL_BODY_SIZE 24 + +#define RPC_REPLY_STATUS_SIZE 4 + +#define RPC_AUTH_FLAVOUR_N_LENGTH_SIZE 8 + +#define RPC_ACCEPT_STATUS_LEN 4 + +struct rpc_transport_ops; +typedef struct rpc_transport rpc_transport_t; + +#include <glusterfs/dict.h> +#include <glusterfs/compat.h> +#include <glusterfs/async.h> +#include "rpcsvc-common.h" + +struct peer_info { + // OP-VERSION of clients + uint32_t max_op_version; + uint32_t min_op_version; + struct sockaddr_storage sockaddr; + socklen_t sockaddr_len; + char identifier[UNIX_PATH_MAX]; + // Volume mounted by client + char volname[NAME_MAX]; +}; +typedef struct peer_info peer_info_t; + +typedef enum msg_type msg_type_t; + +typedef enum { + RPC_TRANSPORT_ACCEPT, /* New client has been accepted */ + RPC_TRANSPORT_DISCONNECT, /* Connection is disconnected */ + RPC_TRANSPORT_CLEANUP, /* connection is about to be freed */ + /*RPC_TRANSPORT_READ,*/ /* An event used to enable rpcsvc to instruct + * transport the number of bytes to read. + * This helps in reading large msgs, wherein + * the rpc actors might decide to place the + * actor's payload in new iobufs separate + * from the rpc header, proghdr and + * authentication information. glusterfs/nfs + * read and write actors are few examples + * that might beniefit from this. While + * reading a single msg, this event may be + * delivered more than once. + */ + RPC_TRANSPORT_MAP_XID_REQUEST, /* receiver of this event should send + * the prognum and procnum corresponding + * to xid. + */ + RPC_TRANSPORT_MSG_RECEIVED, /* Complete rpc msg has been read */ + RPC_TRANSPORT_CONNECT, /* client is connected to server */ + RPC_TRANSPORT_MSG_SENT, + RPC_TRANSPORT_EVENT_THREAD_DIED /* event-thread has died */ +} rpc_transport_event_t; + +struct rpc_transport_msg { + struct iovec *rpchdr; + struct iovec *proghdr; + int rpchdrcount; + int proghdrcount; + struct iovec *progpayload; + struct iobref *iobref; + int progpayloadcount; +}; +typedef struct rpc_transport_msg rpc_transport_msg_t; + +struct rpc_transport_rsp { + struct iovec *rsphdr; + struct iovec *rsp_payload; + int rsphdr_count; + int rsp_payload_count; + struct iobref *rsp_iobref; +}; +typedef struct rpc_transport_rsp rpc_transport_rsp_t; + +struct rpc_transport_req { + struct rpc_req *rpc_req; + rpc_transport_msg_t msg; + rpc_transport_rsp_t rsp; +}; +typedef struct rpc_transport_req rpc_transport_req_t; + +struct rpc_transport_reply { + void *private; + rpc_transport_msg_t msg; +}; +typedef struct rpc_transport_reply rpc_transport_reply_t; + +struct rpc_transport_data { + union { + rpc_transport_req_t req; + rpc_transport_reply_t reply; + } data; + char is_request; +}; +typedef struct rpc_transport_data rpc_transport_data_t; + +/* FIXME: prognum, procnum and progver are already present in + * rpc_request, hence these should be removed from request_info + */ +struct rpc_request_info { + int prognum; + int progver; + void *rpc_req; /* struct rpc_req */ + rpc_transport_rsp_t rsp; + int procnum; + uint32_t xid; +}; +typedef struct rpc_request_info rpc_request_info_t; + +typedef int (*rpc_transport_notify_t)(rpc_transport_t *, void *mydata, + rpc_transport_event_t, void *data, ...); + +struct rpc_transport { + struct rpc_transport_ops *ops; + rpc_transport_t *listener; /* listener transport to which + * request for creation of this + * transport came from. valid only + * on server process. + */ + + void *private; + struct _client *xl_private; + void *xl; /* Used for THIS */ + void *mydata; + pthread_mutex_t lock; + gf_atomic_t refcount; + glusterfs_ctx_t *ctx; + dict_t *options; + char *name; + void *dnscache; + void *drc_client; + data_t *buf; + int32_t (*init)(rpc_transport_t *this); + void (*fini)(rpc_transport_t *this); + int (*reconfigure)(rpc_transport_t *this, dict_t *options); + rpc_transport_notify_t notify; + void *notify_data; + peer_info_t peerinfo; + peer_info_t myinfo; + + uint64_t total_bytes_read; + uint64_t total_bytes_write; + uint32_t xid; /* RPC/XID used for callbacks */ + int32_t outstanding_rpc_count; + + struct list_head list; + void *dl_handle; /* handle of dlopen() */ + char *ssl_name; + dict_t *clnt_options; /* store options received from + * client */ + gf_atomic_t disconnect_progress; + int bind_insecure; + /* connect_failed: saves the connect() syscall status as socket_t + * member holding connect() status can't be accessed by higher gfapi + * layer or in client management notification handler functions + */ + gf_boolean_t connect_failed; + char notify_poller_death; + char poller_death_accept; +}; + +struct rpc_transport_pollin { + struct rpc_transport *trans; + void *private; + struct iobref *iobref; + struct iovec vector[MAX_IOVEC]; + gf_async_t async; + int count; + char is_reply; + char vectored; +}; +typedef struct rpc_transport_pollin rpc_transport_pollin_t; + +struct rpc_transport_ops { + /* no need of receive op, msg will be delivered through an event + * notification + */ + int32_t (*submit_request)(rpc_transport_t *this, rpc_transport_req_t *req); + int32_t (*submit_reply)(rpc_transport_t *this, + rpc_transport_reply_t *reply); + int32_t (*connect)(rpc_transport_t *this, int port); + int32_t (*listen)(rpc_transport_t *this); + int32_t (*disconnect)(rpc_transport_t *this, gf_boolean_t wait); + int32_t (*get_peername)(rpc_transport_t *this, char *hostname, int hostlen); + int32_t (*get_peeraddr)(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, socklen_t sasize); + int32_t (*get_myname)(rpc_transport_t *this, char *hostname, int hostlen); + int32_t (*get_myaddr)(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, socklen_t sasize); + int32_t (*throttle)(rpc_transport_t *this, gf_boolean_t onoff); +}; + +int32_t +rpc_transport_count(const char *transport_type); + +int32_t +rpc_transport_listen(rpc_transport_t *this); + +int32_t +rpc_transport_connect(rpc_transport_t *this, int port); + +int32_t +rpc_transport_disconnect(rpc_transport_t *this, gf_boolean_t wait); + +int32_t +rpc_transport_notify(rpc_transport_t *this, rpc_transport_event_t event, + void *data, ...); + +int32_t +rpc_transport_submit_request(rpc_transport_t *this, rpc_transport_req_t *req); + +int32_t +rpc_transport_submit_reply(rpc_transport_t *this, rpc_transport_reply_t *reply); + +rpc_transport_t * +rpc_transport_load(glusterfs_ctx_t *ctx, dict_t *options, char *name); + +rpc_transport_t * +rpc_transport_ref(rpc_transport_t *trans); + +int32_t +rpc_transport_unref(rpc_transport_t *trans); + +int +rpc_transport_register_notify(rpc_transport_t *trans, rpc_transport_notify_t, + void *mydata); + +int32_t +rpc_transport_get_peername(rpc_transport_t *this, char *hostname, int hostlen); + +int32_t +rpc_transport_get_peeraddr(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, size_t salen); + +int32_t +rpc_transport_get_myaddr(rpc_transport_t *this, char *peeraddr, int addrlen, + struct sockaddr_storage *sa, size_t salen); + +int +rpc_transport_throttle(rpc_transport_t *this, gf_boolean_t onoff); + +rpc_transport_pollin_t * +rpc_transport_pollin_alloc(rpc_transport_t *this, struct iovec *vector, + int count, struct iobuf *hdr_iobuf, + struct iobref *iobref, void *private); +void +rpc_transport_pollin_destroy(rpc_transport_pollin_t *pollin); + +int +rpc_transport_keepalive_options_set(dict_t *options, int32_t interval, + int32_t time, int32_t timeout); + +int +rpc_transport_unix_options_build(dict_t *options, char *filepath, + int frame_timeout); + +int +rpc_transport_inet_options_build(dict_t *options, const char *hostname, + int port, char *af); + +void +rpc_transport_cleanup(rpc_transport_t *); +#endif /* __RPC_TRANSPORT_H__ */ diff --git a/rpc/rpc-lib/src/rpcsvc-auth.c b/rpc/rpc-lib/src/rpcsvc-auth.c new file mode 100644 index 00000000000..8e76b4188bb --- /dev/null +++ b/rpc/rpc-lib/src/rpcsvc-auth.c @@ -0,0 +1,561 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#include <glusterfs/dict.h> + +extern rpcsvc_auth_t * +rpcsvc_auth_null_init(rpcsvc_t *svc, dict_t *options); + +extern rpcsvc_auth_t * +rpcsvc_auth_unix_init(rpcsvc_t *svc, dict_t *options); + +extern rpcsvc_auth_t * +rpcsvc_auth_glusterfs_init(rpcsvc_t *svc, dict_t *options); +extern rpcsvc_auth_t * +rpcsvc_auth_glusterfs_v2_init(rpcsvc_t *svc, dict_t *options); +extern rpcsvc_auth_t * +rpcsvc_auth_glusterfs_v3_init(rpcsvc_t *svc, dict_t *options); + +int +rpcsvc_auth_add_initer(struct list_head *list, char *idfier, + rpcsvc_auth_initer_t init) +{ + struct rpcsvc_auth_list *new = NULL; + + if ((!list) || (!init) || (!idfier)) + return -1; + + new = GF_CALLOC(1, sizeof(*new), gf_common_mt_rpcsvc_auth_list); + if (!new) { + return -1; + } + + new->init = init; + strncpy(new->name, idfier, sizeof(new->name) - 1); + INIT_LIST_HEAD(&new->authlist); + list_add_tail(&new->authlist, list); + return 0; +} + +int +rpcsvc_auth_add_initers(rpcsvc_t *svc) +{ + int ret = -1; + + ret = rpcsvc_auth_add_initer( + &svc->authschemes, "auth-glusterfs", + (rpcsvc_auth_initer_t)rpcsvc_auth_glusterfs_init); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_GLUSTERFS"); + goto err; + } + + ret = rpcsvc_auth_add_initer( + &svc->authschemes, "auth-glusterfs-v2", + (rpcsvc_auth_initer_t)rpcsvc_auth_glusterfs_v2_init); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_GLUSTERFS-v2"); + goto err; + } + + ret = rpcsvc_auth_add_initer( + &svc->authschemes, "auth-glusterfs-v3", + (rpcsvc_auth_initer_t)rpcsvc_auth_glusterfs_v3_init); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_GLUSTERFS-v3"); + goto err; + } + + ret = rpcsvc_auth_add_initer(&svc->authschemes, "auth-unix", + (rpcsvc_auth_initer_t)rpcsvc_auth_unix_init); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_UNIX"); + goto err; + } + + ret = rpcsvc_auth_add_initer(&svc->authschemes, "auth-null", + (rpcsvc_auth_initer_t)rpcsvc_auth_null_init); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_NULL"); + goto err; + } + + ret = 0; +err: + return ret; +} + +int +rpcsvc_auth_init_auth(rpcsvc_t *svc, dict_t *options, + struct rpcsvc_auth_list *authitem) +{ + int ret = -1; + + if ((!svc) || (!options) || (!authitem)) + return -1; + + if (!authitem->init) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "No init function defined"); + ret = -1; + goto err; + } + + authitem->auth = authitem->init(svc, options); + if (!authitem->auth) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Registration of auth failed:" + " %s", + authitem->name); + ret = -1; + goto err; + } + + authitem->enable = 1; + gf_log(GF_RPCSVC, GF_LOG_TRACE, "Authentication enabled: %s", + authitem->auth->authname); + + ret = 0; +err: + return ret; +} + +int +rpcsvc_auth_init_auths(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + + if (!svc) + return -1; + + if (list_empty(&svc->authschemes)) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "No authentication!"); + ret = 0; + goto err; + } + + /* If auth null and sys are not disabled by the user, we must enable + * it by default. This is a globally default rule, the user is still + * allowed to disable the two for particular subvolumes. + */ + if (!dict_get(options, "rpc-auth.auth-null")) { + ret = dict_set_str(options, "rpc-auth.auth-null", "on"); + if (ret) + gf_log("rpc-auth", GF_LOG_DEBUG, "dict_set failed for 'auth-nill'"); + } + + if (!dict_get(options, "rpc-auth.auth-unix")) { + ret = dict_set_str(options, "rpc-auth.auth-unix", "on"); + if (ret) + gf_log("rpc-auth", GF_LOG_DEBUG, "dict_set failed for 'auth-unix'"); + } + + if (!dict_get(options, "rpc-auth.auth-glusterfs")) { + ret = dict_set_str(options, "rpc-auth.auth-glusterfs", "on"); + if (ret) + gf_log("rpc-auth", GF_LOG_DEBUG, "dict_set failed for 'auth-unix'"); + } + + list_for_each_entry_safe(auth, tmp, &svc->authschemes, authlist) + { + ret = rpcsvc_auth_init_auth(svc, options, auth); + if (ret == -1) + goto err; + } + + ret = 0; +err: + return ret; +} + +int +rpcsvc_set_addr_namelookup(rpcsvc_t *svc, dict_t *options) +{ + int ret; + static char *addrlookup_key = "rpc-auth.addr.namelookup"; + + if (!svc || !options) + return (-1); + + /* By default it's disabled */ + ret = dict_get_str_boolean(options, addrlookup_key, _gf_false); + if (ret < 0) { + svc->addr_namelookup = _gf_false; + } else { + svc->addr_namelookup = ret; + } + + if (svc->addr_namelookup) + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "Addr-Name lookup enabled"); + + return (0); +} + +int +rpcsvc_set_allow_insecure(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + char *allow_insecure_str = NULL; + gf_boolean_t is_allow_insecure = _gf_false; + + GF_ASSERT(svc); + GF_ASSERT(options); + + ret = dict_get_str(options, "rpc-auth-allow-insecure", &allow_insecure_str); + if (0 == ret) { + ret = gf_string2boolean(allow_insecure_str, &is_allow_insecure); + if (0 == ret) { + if (_gf_true == is_allow_insecure) + svc->allow_insecure = 1; + else + svc->allow_insecure = 0; + } + } else { + /* By default set allow-insecure to true */ + svc->allow_insecure = 1; + + /* setting in options for the sake of functions that look + * configuration params for allow insecure, eg: gf_auth + */ + ret = dict_set_str(options, "rpc-auth-allow-insecure", "on"); + if (ret < 0) + gf_log("rpc-auth", GF_LOG_DEBUG, + "dict_set failed for 'allow-insecure'"); + } + + return ret; +} + +int +rpcsvc_set_root_squash(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + uid_t anonuid = -1; + gid_t anongid = -1; + + GF_ASSERT(svc); + GF_ASSERT(options); + + ret = dict_get_str_boolean(options, "root-squash", 0); + if (ret != -1) + svc->root_squash = ret; + else + svc->root_squash = _gf_false; + + ret = dict_get_uint32(options, "anonuid", &anonuid); + if (!ret) + svc->anonuid = anonuid; + else + svc->anonuid = RPC_NOBODY_UID; + + ret = dict_get_uint32(options, "anongid", &anongid); + if (!ret) + svc->anongid = anongid; + else + svc->anongid = RPC_NOBODY_GID; + + if (svc->root_squash) + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "root squashing enabled " + "(uid=%d, gid=%d)", + svc->anonuid, svc->anongid); + + return 0; +} + +int +rpcsvc_set_all_squash(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + + uid_t anonuid = -1; + gid_t anongid = -1; + + GF_ASSERT(svc); + GF_ASSERT(options); + + ret = dict_get_str_boolean(options, "all-squash", 0); + if (ret != -1) + svc->all_squash = ret; + else + svc->all_squash = _gf_false; + + ret = dict_get_uint32(options, "anonuid", &anonuid); + if (!ret) + svc->anonuid = anonuid; + else + svc->anonuid = RPC_NOBODY_UID; + + ret = dict_get_uint32(options, "anongid", &anongid); + if (!ret) + svc->anongid = anongid; + else + svc->anongid = RPC_NOBODY_GID; + + if (svc->all_squash) + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "all squashing enabled " + "(uid=%d, gid=%d)", + svc->anonuid, svc->anongid); + + return 0; +} + +int +rpcsvc_auth_init(rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + + if ((!svc) || (!options)) + return -1; + + (void)rpcsvc_set_allow_insecure(svc, options); + (void)rpcsvc_set_root_squash(svc, options); + (void)rpcsvc_set_all_squash(svc, options); + (void)rpcsvc_set_addr_namelookup(svc, options); + ret = rpcsvc_auth_add_initers(svc); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to add initers"); + goto out; + } + + ret = rpcsvc_auth_init_auths(svc, options); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to init auth schemes"); + goto out; + } + +out: + return ret; +} + +int +rpcsvc_auth_reconf(rpcsvc_t *svc, dict_t *options) +{ + int ret = 0; + + if ((!svc) || (!options)) + return (-1); + + ret = rpcsvc_set_allow_insecure(svc, options); + if (ret) + return (-1); + + ret = rpcsvc_set_root_squash(svc, options); + if (ret) + return (-1); + + ret = rpcsvc_set_all_squash(svc, options); + if (ret) + return (-1); + + return rpcsvc_set_addr_namelookup(svc, options); +} + +rpcsvc_auth_t * +__rpcsvc_auth_get_handler(rpcsvc_request_t *req) +{ + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + rpcsvc_t *svc = NULL; + + if (!req) + return NULL; + + svc = req->svc; + if (!svc) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "!svc"); + goto err; + } + + if (list_empty(&svc->authschemes)) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "No authentication!"); + goto err; + } + + list_for_each_entry_safe(auth, tmp, &svc->authschemes, authlist) + { + if (!auth->enable) + continue; + if (auth->auth->authnum == req->cred.flavour) + goto err; + } + + auth = NULL; +err: + if (auth) + return auth->auth; + else + return NULL; +} + +rpcsvc_auth_t * +rpcsvc_auth_get_handler(rpcsvc_request_t *req) +{ + rpcsvc_auth_t *auth = NULL; + + auth = __rpcsvc_auth_get_handler(req); + if (auth) + goto ret; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, "No auth handler: %d", req->cred.flavour); + + /* The requested scheme was not available so fall back the to one + * scheme that will always be present. + */ + req->cred.flavour = AUTH_NULL; + req->verf.flavour = AUTH_NULL; + auth = __rpcsvc_auth_get_handler(req); +ret: + return auth; +} + +int +rpcsvc_auth_request_init(rpcsvc_request_t *req, struct rpc_msg *callmsg) +{ + int32_t ret = 0; + rpcsvc_auth_t *auth = NULL; + + if (!req || !callmsg) { + ret = -1; + goto err; + } + + req->cred.flavour = rpc_call_cred_flavour(callmsg); + req->cred.datalen = rpc_call_cred_len(callmsg); + req->verf.flavour = rpc_call_verf_flavour(callmsg); + req->verf.datalen = rpc_call_verf_len(callmsg); + + auth = rpcsvc_auth_get_handler(req); + if (!auth) { + ret = -1; + goto err; + } + + gf_log(GF_RPCSVC, GF_LOG_TRACE, "Auth handler: %s", auth->authname); + + if (auth->authops->request_init) + ret = auth->authops->request_init(req, auth->authprivate); + + /* reset to auxgidlarge during + unsersialize if necessary */ + req->auxgids = req->auxgidsmall; + req->auxgidlarge = NULL; +err: + return ret; +} + +int +rpcsvc_authenticate(rpcsvc_request_t *req) +{ + int ret = RPCSVC_AUTH_REJECT; + rpcsvc_auth_t *auth = NULL; + int minauth = 0; + + if (!req) + return ret; + + /* FIXME use rpcsvc_request_prog_minauth() */ + minauth = 0; + if (minauth > rpcsvc_request_cred_flavour(req)) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "Auth too weak"); + rpcsvc_request_set_autherr(req, AUTH_TOOWEAK); + goto err; + } + + auth = rpcsvc_auth_get_handler(req); + if (!auth) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "No auth handler found"); + goto err; + } + + if (auth->authops->authenticate) + ret = auth->authops->authenticate(req, auth->authprivate); + +err: + return ret; +} + +int +rpcsvc_auth_array(rpcsvc_t *svc, char *volname, int *autharr, int arrlen) +{ + int count = 0; + int result = RPCSVC_AUTH_REJECT; + char *srchstr = NULL; + int ret = 0; + + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + + if ((!svc) || (!autharr) || (!volname)) + return -1; + + memset(autharr, 0, arrlen * sizeof(int)); + if (list_empty(&svc->authschemes)) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "No authentication!"); + goto err; + } + + list_for_each_entry_safe(auth, tmp, &svc->authschemes, authlist) + { + if (count >= arrlen) + break; + + result = gf_asprintf(&srchstr, "rpc-auth.%s.%s", auth->name, volname); + if (result == -1) { + count = -1; + goto err; + } + + ret = dict_get_str_boolean(svc->options, srchstr, 0xC00FFEE); + GF_FREE(srchstr); + + switch (ret) { + case _gf_true: + autharr[count] = auth->auth->authnum; + ++count; + break; + + default: + /* nothing to do */ + break; + } + } + +err: + return count; +} + +gid_t * +rpcsvc_auth_unix_auxgids(rpcsvc_request_t *req, int *arrlen) +{ + if ((!req) || (!arrlen)) + return NULL; + + /* In case of AUTH_NULL auxgids are not used */ + switch (req->cred.flavour) { + case AUTH_UNIX: + case AUTH_GLUSTERFS: + case AUTH_GLUSTERFS_v2: + case AUTH_GLUSTERFS_v3: + break; + default: + gf_log("rpc", GF_LOG_DEBUG, "auth type not unix or glusterfs"); + return NULL; + } + + *arrlen = req->auxgidcount; + if (*arrlen == 0) + return NULL; + + return &req->auxgids[0]; +} diff --git a/rpc/rpc-lib/src/rpcsvc-common.h b/rpc/rpc-lib/src/rpcsvc-common.h new file mode 100644 index 00000000000..6c4ec49a6ef --- /dev/null +++ b/rpc/rpc-lib/src/rpcsvc-common.h @@ -0,0 +1,113 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _RPCSVC_COMMON_H +#define _RPCSVC_COMMON_H + +#include <pthread.h> +#include <glusterfs/compat.h> +#include <glusterfs/dict.h> + +typedef enum { + RPCSVC_EVENT_ACCEPT, + RPCSVC_EVENT_DISCONNECT, + RPCSVC_EVENT_TRANSPORT_DESTROY, + RPCSVC_EVENT_LISTENER_DEAD, +} rpcsvc_event_t; + +struct rpcsvc_state; + +typedef int (*rpcsvc_notify_t)(struct rpcsvc_state *, void *mydata, + rpcsvc_event_t, void *data); + +struct drc_globals; +typedef struct drc_globals rpcsvc_drc_globals_t; + +/* Contains global state required for all the RPC services. + */ +typedef struct rpcsvc_state { + /* Contains list of (program, version) handlers. + * other options. + */ + + pthread_rwlock_t rpclock; + + /* List of the authentication schemes available. */ + struct list_head authschemes; + + /* Reference to the options */ + dict_t *options; + + uid_t anonuid; + gid_t anongid; + glusterfs_ctx_t *ctx; + + /* list of connections which will listen for incoming connections */ + struct list_head listeners; + + /* list of programs registered with rpcsvc */ + struct list_head programs; + + /* list of notification callbacks */ + struct list_head notify; + int notify_count; + + unsigned int memfactor; + + xlator_t *xl; /* xlator */ + void *mydata; + rpcsvc_notify_t notifyfn; + struct mem_pool *rxpool; + rpcsvc_drc_globals_t *drc; + + /* per-client limit of outstanding rpc requests */ + int outstanding_rpc_limit; + gf_boolean_t addr_namelookup; + /* determine whether throttling is needed, by default OFF */ + gf_boolean_t throttle; + /* Allow insecure ports. */ + gf_boolean_t allow_insecure; + gf_boolean_t register_portmap; + gf_boolean_t root_squash; + gf_boolean_t all_squash; +} rpcsvc_t; + +/* DRC START */ +enum drc_op_type { DRC_NA = 0, DRC_IDEMPOTENT = 1, DRC_NON_IDEMPOTENT = 2 }; +typedef enum drc_op_type drc_op_type_t; + +enum drc_type { DRC_TYPE_NONE = 0, DRC_TYPE_IN_MEMORY = 1 }; +typedef enum drc_type drc_type_t; + +enum drc_lru_factor { + DRC_LRU_5_PC = 20, + DRC_LRU_10_PC = 10, + DRC_LRU_25_PC = 4, + DRC_LRU_50_PC = 2 +}; +typedef enum drc_lru_factor drc_lru_factor_t; + +enum drc_xid_state { DRC_XID_MONOTONOUS = 0, DRC_XID_WRAPPED = 1 }; +typedef enum drc_xid_state drc_xid_state_t; + +enum drc_op_state { DRC_OP_IN_TRANSIT = 0, DRC_OP_CACHED = 1 }; +typedef enum drc_op_state drc_op_state_t; + +enum drc_policy { DRC_LRU = 0 }; +typedef enum drc_policy drc_policy_t; + +/* Default policies for DRC */ +#define DRC_DEFAULT_TYPE DRC_TYPE_IN_MEMORY +#define DRC_DEFAULT_CACHE_SIZE 0x20000 +#define DRC_DEFAULT_LRU_FACTOR DRC_LRU_25_PC + +/* DRC END */ + +#endif /* #ifndef _RPCSVC_COMMON_H */ diff --git a/rpc/rpc-lib/src/rpcsvc.c b/rpc/rpc-lib/src/rpcsvc.c new file mode 100644 index 00000000000..39910d481bf --- /dev/null +++ b/rpc/rpc-lib/src/rpcsvc.c @@ -0,0 +1,3295 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include "rpcsvc.h" +#include "rpc-transport.h" +#include <glusterfs/dict.h> +#include <glusterfs/byte-order.h> +#include <glusterfs/compat-errno.h> +#include <glusterfs/statedump.h> +#include "xdr-rpc.h" +#include <glusterfs/iobuf.h> +#include "xdr-common.h" +#include "xdr-generic.h" +#include "rpc-common-xdr.h" +#include <glusterfs/syncop.h> +#include "rpc-drc.h" +#include "protocol-common.h" + +#include <errno.h> +#include <pthread.h> +#include <stdlib.h> +#include <rpc/rpc.h> +#include <rpc/pmap_clnt.h> +#include <arpa/inet.h> +#include <rpc/xdr.h> +#include <fnmatch.h> +#include <stdarg.h> +#include <stdio.h> +#include <dlfcn.h> + +#ifdef IPV6_DEFAULT +#include <netconfig.h> +#endif + +#include "xdr-rpcclnt.h" +#include <glusterfs/glusterfs-acl.h> + +#ifndef PTHREAD_MUTEX_ADAPTIVE_NP +#define PTHREAD_MUTEX_ADAPTIVE_NP PTHREAD_MUTEX_DEFAULT +#endif + +static struct rpcsvc_program gluster_dump_prog; + +#define rpcsvc_alloc_request(svc, request) \ + do { \ + request = (rpcsvc_request_t *)mem_get((svc)->rxpool); \ + if (request) { \ + memset(request, 0, sizeof(rpcsvc_request_t)); \ + } else { \ + gf_log("rpcsvc", GF_LOG_ERROR, \ + "error getting memory for rpc request"); \ + } \ + } while (0) + +rpcsvc_listener_t * +rpcsvc_get_listener(rpcsvc_t *svc, uint16_t port, rpc_transport_t *trans); + +int +rpcsvc_notify(rpc_transport_t *trans, void *mydata, rpc_transport_event_t event, + void *data, ...); +void * +rpcsvc_request_handler(void *arg); + +static int +rpcsvc_match_subnet_v4(const char *addrtok, const char *ipaddr); + +static void +rpcsvc_toggle_queue_status(rpcsvc_program_t *prog, + rpcsvc_request_queue_t *queue, + unsigned long status[]) +{ + unsigned queue_index = queue - prog->request_queue; + + status[queue_index / __BITS_PER_LONG] ^= (1UL << (queue_index % + __BITS_PER_LONG)); +} + +int +rpcsvc_get_free_queue_index(rpcsvc_program_t *prog) +{ + unsigned i, j = 0; + + for (i = 0; i < EVENT_MAX_THREADS / __BITS_PER_LONG; i++) + if (prog->request_queue_status[i] != ULONG_MAX) { + j = __builtin_ctzl(~prog->request_queue_status[i]); + break; + } + + if (i == EVENT_MAX_THREADS / __BITS_PER_LONG) + return -1; + + prog->request_queue_status[i] |= (1UL << j); + return i * __BITS_PER_LONG + j; +} + +rpcsvc_notify_wrapper_t * +rpcsvc_notify_wrapper_alloc(void) +{ + rpcsvc_notify_wrapper_t *wrapper = NULL; + + wrapper = GF_CALLOC(1, sizeof(*wrapper), gf_common_mt_rpcsvc_wrapper_t); + if (!wrapper) { + goto out; + } + + INIT_LIST_HEAD(&wrapper->list); +out: + return wrapper; +} + +void +rpcsvc_listener_destroy(rpcsvc_listener_t *listener) +{ + rpcsvc_t *svc = NULL; + + if (!listener) { + goto out; + } + + svc = listener->svc; + if (!svc) { + goto listener_free; + } + + pthread_rwlock_wrlock(&svc->rpclock); + { + list_del_init(&listener->list); + } + pthread_rwlock_unlock(&svc->rpclock); + +listener_free: + GF_FREE(listener); +out: + return; +} + +rpcsvc_vector_sizer +rpcsvc_get_program_vector_sizer(rpcsvc_t *svc, uint32_t prognum, + uint32_t progver, int procnum) +{ + rpcsvc_program_t *program = NULL; + char found = 0; + + if (!svc) + return NULL; + + pthread_rwlock_rdlock(&svc->rpclock); + { + /* Find the matching RPC program from registered list */ + list_for_each_entry(program, &svc->programs, program) + { + if ((program->prognum == prognum) && + (program->progver == progver)) { + found = 1; + break; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + + if (found) { + /* Make sure the requested procnum is supported by RPC prog */ + if ((procnum < 0) || (procnum >= program->numactors)) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "RPC procedure %d not available for Program %s", procnum, + program->progname); + return NULL; + } + + /* SUCCESS: Supported procedure */ + return program->actors[procnum].vector_sizer; + } + + return NULL; /* FAIL */ +} + +gf_boolean_t +rpcsvc_can_outstanding_req_be_ignored(rpcsvc_request_t *req) +{ + /* + * If outstanding_rpc_limit is reached because of blocked locks and + * throttling is attempted then no unlock requests will be received. So + * the outstanding request count will never change i.e. it will always + * be equal to the limit. This also leads to ping timer expiry on + * client. + */ + + /* + * This is a hack and a necessity until grantedlock == fop completion. + * Ideally if we get a blocking lock request which cannot be granted + * right now, we should unwind the fop saying “request registered, will + * notify you when granted”, which is very hard to implement at the + * moment. Until we bring in such mechanism, we will need to live with + * not rate-limiting INODELK/ENTRYLK/LK fops + */ + + if ((req->prognum == GLUSTER_FOP_PROGRAM) && + (req->progver == GLUSTER_FOP_VERSION)) { + if ((req->procnum == GFS3_OP_INODELK) || + (req->procnum == GFS3_OP_FINODELK) || + (req->procnum == GFS3_OP_ENTRYLK) || + (req->procnum == GFS3_OP_FENTRYLK) || (req->procnum == GFS3_OP_LK)) + return _gf_true; + } + return _gf_false; +} + +int +rpcsvc_request_outstanding(rpcsvc_request_t *req, int delta) +{ + int ret = -1; + int old_count = 0; + int new_count = 0; + int limit = 0; + gf_boolean_t throttle = _gf_false; + + if (!req) + goto out; + + throttle = rpcsvc_get_throttle(req->svc); + if (!throttle) { + ret = 0; + goto out; + } + + if (rpcsvc_can_outstanding_req_be_ignored(req)) { + ret = 0; + goto out; + } + + pthread_mutex_lock(&req->trans->lock); + { + limit = req->svc->outstanding_rpc_limit; + if (!limit) + goto unlock; + + old_count = req->trans->outstanding_rpc_count; + req->trans->outstanding_rpc_count += delta; + new_count = req->trans->outstanding_rpc_count; + + if (old_count <= limit && new_count > limit) + ret = rpc_transport_throttle(req->trans, _gf_true); + + if (old_count > limit && new_count <= limit) + ret = rpc_transport_throttle(req->trans, _gf_false); + } +unlock: + pthread_mutex_unlock(&req->trans->lock); + +out: + return ret; +} + +/* This needs to change to returning errors, since + * we need to return RPC specific error messages when some + * of the pointers below are NULL. + */ +rpcsvc_actor_t * +rpcsvc_program_actor(rpcsvc_request_t *req) +{ + rpcsvc_program_t *program = NULL; + int err = SYSTEM_ERR; + rpcsvc_actor_t *actor = NULL; + rpcsvc_t *svc = NULL; + char found = 0; + char *peername = NULL; + + if (!req) + goto err; + + svc = req->svc; + peername = req->trans->peerinfo.identifier; + pthread_rwlock_rdlock(&svc->rpclock); + { + list_for_each_entry(program, &svc->programs, program) + { + if (program->prognum == req->prognum) { + err = PROG_MISMATCH; + } + + if ((program->prognum == req->prognum) && + (program->progver == req->progver)) { + found = 1; + break; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + + if (!found) { + if (err != PROG_MISMATCH) { + /* log in DEBUG when nfs clients try to see if + * ACL requests are accepted by nfs server + */ + gf_log( + GF_RPCSVC, + (req->prognum == ACL_PROGRAM) ? GF_LOG_DEBUG : GF_LOG_WARNING, + "RPC program not available (req %u %u) for %s", req->prognum, + req->progver, peername); + err = PROG_UNAVAIL; + goto err; + } + + gf_log(GF_RPCSVC, GF_LOG_WARNING, + "RPC program version not available (req %u %u) for %s", + req->prognum, req->progver, peername); + goto err; + } + req->prog = program; + if (!program->actors) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, + "RPC Actor not found for program %s %d for %s", + program->progname, program->prognum, peername); + err = SYSTEM_ERR; + goto err; + } + + if ((req->procnum < 0) || (req->procnum >= program->numactors)) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "RPC Program procedure not" + " available for procedure %d in %s for %s", + req->procnum, program->progname, peername); + err = PROC_UNAVAIL; + goto err; + } + + actor = &program->actors[req->procnum]; + if (!actor->actor) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "RPC Program procedure not" + " available for procedure %d in %s for %s", + req->procnum, program->progname, peername); + err = PROC_UNAVAIL; + actor = NULL; + goto err; + } + + if (svc->xl->ctx->measure_latency) { + timespec_now(&req->begin); + } + + req->ownthread = program->ownthread; + req->synctask = program->synctask; + + err = SUCCESS; + gf_log(GF_RPCSVC, GF_LOG_TRACE, "Actor found: %s - %s for %s", + program->progname, actor->procname, peername); +err: + if (req) + req->rpc_err = err; + + return actor; +} + +/* this procedure can only pass 4 arguments to registered notifyfn. To send more + * arguments call wrapper->notify directly. + */ +static void +rpcsvc_program_notify(rpcsvc_listener_t *listener, rpcsvc_event_t event, + void *data) +{ + rpcsvc_notify_wrapper_t *wrapper = NULL; + + if (!listener) { + goto out; + } + + list_for_each_entry(wrapper, &listener->svc->notify, list) + { + if (wrapper->notify) { + wrapper->notify(listener->svc, wrapper->data, event, data); + } + } + +out: + return; +} + +static int +rpcsvc_accept(rpcsvc_t *svc, rpc_transport_t *listen_trans, + rpc_transport_t *new_trans) +{ + rpcsvc_listener_t *listener = NULL; + int32_t ret = -1; + + listener = rpcsvc_get_listener(svc, -1, listen_trans); + if (listener == NULL) { + goto out; + } + + rpcsvc_program_notify(listener, RPCSVC_EVENT_ACCEPT, new_trans); + ret = 0; +out: + return ret; +} + +void +rpcsvc_request_destroy(rpcsvc_request_t *req) +{ + if (!req) { + goto out; + } + + if (req->iobref) { + iobref_unref(req->iobref); + } + + /* This marks the "end" of an RPC request. Reply is + completely written to the socket and is on the way + to the client. It is time to decrement the + outstanding request counter by 1. + */ + if (req->prognum) // Only for initialized requests + rpcsvc_request_outstanding(req, -1); + + rpc_transport_unref(req->trans); + + GF_FREE(req->auxgidlarge); + + mem_put(req); + +out: + return; +} + +rpcsvc_request_t * +rpcsvc_request_init(rpcsvc_t *svc, rpc_transport_t *trans, + struct rpc_msg *callmsg, struct iovec progmsg, + rpc_transport_pollin_t *msg, rpcsvc_request_t *req) +{ + int i = 0; + + if ((!trans) || (!callmsg) || (!req) || (!msg)) + return NULL; + + /* We start a RPC request as always denied. */ + req->rpc_status = MSG_DENIED; + req->xid = rpc_call_xid(callmsg); + req->prognum = rpc_call_program(callmsg); + req->progver = rpc_call_progver(callmsg); + req->procnum = rpc_call_progproc(callmsg); + req->trans = rpc_transport_ref(trans); + req->count = msg->count; + req->msg[0] = progmsg; + req->iobref = iobref_ref(msg->iobref); + if (msg->vectored) { + /* msg->vector[MAX_IOVEC] is defined in structure. prevent a + out of bound access */ + for (i = 1; i < min(msg->count, MAX_IOVEC); i++) { + req->msg[i] = msg->vector[i]; + } + } + + req->svc = svc; + req->trans_private = msg->private; + + INIT_LIST_HEAD(&req->txlist); + INIT_LIST_HEAD(&req->request_list); + req->payloadsize = 0; + + /* By this time, the data bytes for the auth scheme would have already + * been copied into the required sections of the req structure, + * we just need to fill in the meta-data about it now. + */ + rpcsvc_auth_request_init(req, callmsg); + return req; +} + +rpcsvc_request_t * +rpcsvc_request_create(rpcsvc_t *svc, rpc_transport_t *trans, + rpc_transport_pollin_t *msg) +{ + char *msgbuf = NULL; + struct rpc_msg rpcmsg; + struct iovec progmsg; /* RPC Program payload */ + rpcsvc_request_t *req = NULL; + size_t msglen = 0; + int ret = -1; + + if (!svc || !trans || !svc->rxpool) + return NULL; + + /* We need to allocate the request before actually calling + * rpcsvc_request_init on the request so that we, can fill the auth + * data directly into the request structure from the message iobuf. + * This avoids a need to keep a temp buffer into which the auth data + * would've been copied otherwise. + */ + rpcsvc_alloc_request(svc, req); + if (!req) { + goto err; + } + + msgbuf = msg->vector[0].iov_base; + msglen = msg->vector[0].iov_len; + + ret = xdr_to_rpc_call(msgbuf, msglen, &rpcmsg, &progmsg, req->cred.authdata, + req->verf.authdata); + + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "RPC call decoding failed"); + rpcsvc_request_seterr(req, GARBAGE_ARGS); + req->trans = rpc_transport_ref(trans); + req->svc = svc; + goto err; + } + + ret = -1; + rpcsvc_request_init(svc, trans, &rpcmsg, progmsg, msg, req); + + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "received rpc-message " + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION + ", Program: %" GF_PRI_RPC_PROG_ID + ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC + ") " + "from rpc-transport (%s)", + rpc_call_xid(&rpcmsg), rpc_call_rpcvers(&rpcmsg), + rpc_call_program(&rpcmsg), rpc_call_progver(&rpcmsg), + rpc_call_progproc(&rpcmsg), trans->name); + + /* We just received a new request from the wire. Account for + it in the outsanding request counter to make sure we don't + ingest too many concurrent requests from the same client. + */ + if (req->prognum) // Only for initialized requests + ret = rpcsvc_request_outstanding(req, +1); + + if (rpc_call_rpcvers(&rpcmsg) != 2) { + /* LOG- TODO: print rpc version, also print the peerinfo + from transport */ + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "RPC version not supported " + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION + ", Program: %" GF_PRI_RPC_PROG_ID + ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC + ") " + "from trans (%s)", + rpc_call_xid(&rpcmsg), rpc_call_rpcvers(&rpcmsg), + rpc_call_program(&rpcmsg), rpc_call_progver(&rpcmsg), + rpc_call_progproc(&rpcmsg), trans->name); + rpcsvc_request_seterr(req, RPC_MISMATCH); + goto err; + } + + ret = rpcsvc_authenticate(req); + if (ret == RPCSVC_AUTH_REJECT) { + /* No need to set auth_err, that is the responsibility of + * the authentication handler since only that know what exact + * error happened. + */ + rpcsvc_request_seterr(req, AUTH_ERROR); + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "auth failed on request. " + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION + ", Program: %" GF_PRI_RPC_PROG_ID + ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC + ") " + "from trans (%s)", + rpc_call_xid(&rpcmsg), rpc_call_rpcvers(&rpcmsg), + rpc_call_program(&rpcmsg), rpc_call_progver(&rpcmsg), + rpc_call_progproc(&rpcmsg), trans->name); + ret = -1; + goto err; + } + + /* If the error is not RPC_MISMATCH, we consider the call as accepted + * since we are not handling authentication failures for now. + */ + req->rpc_status = MSG_ACCEPTED; + req->reply = NULL; + ret = 0; +err: + if (ret == -1) { + ret = rpcsvc_error_reply(req); + if (ret) + gf_log("rpcsvc", GF_LOG_WARNING, "failed to queue error reply"); + req = NULL; + } + + return req; +} + +int +rpcsvc_check_and_reply_error(int ret, call_frame_t *frame, void *opaque) +{ + rpcsvc_request_t *req = NULL; + + req = opaque; + + if (ret) + gf_log("rpcsvc", GF_LOG_ERROR, + "rpc actor (%d:%d:%d) failed to complete successfully", + req->prognum, req->progver, req->procnum); + + if (ret == RPCSVC_ACTOR_ERROR) { + ret = rpcsvc_error_reply(req); + if (ret) + gf_log("rpcsvc", GF_LOG_WARNING, "failed to queue error reply"); + } + + return 0; +} + +void +rpcsvc_queue_event_thread_death(rpcsvc_t *svc, rpcsvc_program_t *prog, int gen) +{ + rpcsvc_request_queue_t *queue = NULL; + int num = 0; + void *value = NULL; + rpcsvc_request_t *req = NULL; + char empty = 0; + + value = pthread_getspecific(prog->req_queue_key); + if (value == NULL) { + return; + } + + num = ((unsigned long)value) - 1; + + queue = &prog->request_queue[num]; + + if (queue->gen == gen) { + /* duplicate event */ + gf_log(GF_RPCSVC, GF_LOG_INFO, + "not queuing duplicate event thread death. " + "queue %d program %s", + num, prog->progname); + return; + } + + rpcsvc_alloc_request(svc, req); + req->prognum = RPCSVC_INFRA_PROGRAM; + req->procnum = RPCSVC_PROC_EVENT_THREAD_DEATH; + gf_log(GF_RPCSVC, GF_LOG_INFO, + "queuing event thread death request to queue %d of program %s", num, + prog->progname); + + pthread_mutex_lock(&queue->queue_lock); + { + empty = list_empty(&queue->request_queue); + + list_add_tail(&req->request_list, &queue->request_queue); + queue->gen = gen; + + if (empty && queue->waiting) + pthread_cond_signal(&queue->queue_cond); + } + pthread_mutex_unlock(&queue->queue_lock); + + return; +} + +int +rpcsvc_handle_event_thread_death(rpcsvc_t *svc, rpc_transport_t *trans, int gen) +{ + rpcsvc_program_t *prog = NULL; + + pthread_rwlock_rdlock(&svc->rpclock); + { + list_for_each_entry(prog, &svc->programs, program) + { + if (prog->ownthread) + rpcsvc_queue_event_thread_death(svc, prog, gen); + } + } + pthread_rwlock_unlock(&svc->rpclock); + + return 0; +} + +int +rpcsvc_handle_rpc_call(rpcsvc_t *svc, rpc_transport_t *trans, + rpc_transport_pollin_t *msg) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_actor actor_fn = NULL; + rpcsvc_request_t *req = NULL; + int ret = -1; + uint16_t port = 0; + gf_boolean_t is_unix = _gf_false, empty = _gf_false; + gf_boolean_t unprivileged = _gf_false, spawn_request_handler = 0; + drc_cached_op_t *reply = NULL; + rpcsvc_drc_globals_t *drc = NULL; + rpcsvc_request_queue_t *queue = NULL; + long num = 0; + void *value = NULL; + + if (!trans || !svc) + return -1; + + switch (trans->peerinfo.sockaddr.ss_family) { + case AF_INET: + port = ((struct sockaddr_in *)&trans->peerinfo.sockaddr)->sin_port; + break; + + case AF_INET6: + port = ((struct sockaddr_in6 *)&trans->peerinfo.sockaddr) + ->sin6_port; + break; + case AF_UNIX: + is_unix = _gf_true; + break; + default: + gf_log(GF_RPCSVC, GF_LOG_ERROR, "invalid address family (%d)", + trans->peerinfo.sockaddr.ss_family); + return -1; + } + + if (is_unix == _gf_false) { + port = ntohs(port); + + gf_log("rpcsvc", GF_LOG_TRACE, "Client port: %d", (int)port); + + if (port >= 1024) + unprivileged = _gf_true; + } + + req = rpcsvc_request_create(svc, trans, msg); + if (!req) + goto out; + + if (!rpcsvc_request_accepted(req)) + goto err_reply; + + actor = rpcsvc_program_actor(req); + if (!actor) + goto err_reply; + + if (0 == svc->allow_insecure && unprivileged && !actor->unprivileged) { + /* Non-privileged user, fail request */ + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Request received from non-" + "privileged port. Failing request for %s.", + req->trans->peerinfo.identifier); + req->rpc_status = MSG_DENIED; + req->rpc_err = AUTH_ERROR; + req->auth_err = RPCSVC_AUTH_REJECT; + goto err_reply; + } + + /* DRC */ + if (rpcsvc_need_drc(req)) { + drc = req->svc->drc; + + LOCK(&drc->lock); + { + reply = rpcsvc_drc_lookup(req); + + /* retransmission of completed request, send cached reply */ + if (reply && reply->state == DRC_OP_CACHED) { + gf_log(GF_RPCSVC, GF_LOG_INFO, + "duplicate request:" + " XID: 0x%x", + req->xid); + ret = rpcsvc_send_cached_reply(req, reply); + drc->cache_hits++; + UNLOCK(&drc->lock); + goto out; + + } /* retransmitted request, original op in transit, drop it */ + else if (reply && reply->state == DRC_OP_IN_TRANSIT) { + gf_log(GF_RPCSVC, GF_LOG_INFO, + "op in transit," + " discarding. XID: 0x%x", + req->xid); + ret = 0; + drc->intransit_hits++; + rpcsvc_request_destroy(req); + UNLOCK(&drc->lock); + goto out; + + } /* fresh request, cache it as in-transit and proceed */ + else { + ret = rpcsvc_cache_request(req); + } + } + UNLOCK(&drc->lock); + } + + if (req->rpc_err == SUCCESS) { + /* Before going to xlator code, set the THIS properly */ + THIS = svc->xl; + + actor_fn = actor->actor; + + if (!actor_fn) { + rpcsvc_request_seterr(req, PROC_UNAVAIL); + /* LOG TODO: print more info about procnum, + prognum etc, also print transport info */ + gf_log(GF_RPCSVC, GF_LOG_ERROR, "No vectored handler present"); + ret = RPCSVC_ACTOR_ERROR; + goto err_reply; + } + + if (req->synctask) { + ret = synctask_new(THIS->ctx->env, (synctask_fn_t)actor_fn, + rpcsvc_check_and_reply_error, NULL, req); + } else if (req->ownthread) { + value = pthread_getspecific(req->prog->req_queue_key); + if (value == NULL) { + pthread_mutex_lock(&req->prog->thr_lock); + { + num = rpcsvc_get_free_queue_index(req->prog); + if (num != -1) { + num++; + value = (void *)num; + ret = pthread_setspecific(req->prog->req_queue_key, + value); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, + "setting request queue in TLS failed"); + rpcsvc_toggle_queue_status( + req->prog, &req->prog->request_queue[num - 1], + req->prog->request_queue_status); + num = -1; + } else { + spawn_request_handler = 1; + } + } + } + pthread_mutex_unlock(&req->prog->thr_lock); + } + + if (num == -1) + goto noqueue; + + num = ((unsigned long)value) - 1; + + queue = &req->prog->request_queue[num]; + + if (spawn_request_handler) { + ret = gf_thread_create(&queue->thread, NULL, + rpcsvc_request_handler, queue, + "rpcrqhnd"); + if (!ret) { + gf_log(GF_RPCSVC, GF_LOG_INFO, + "spawned a request handler thread for queue %d", + (int)num); + + req->prog->threadcount++; + } else { + gf_log( + GF_RPCSVC, GF_LOG_INFO, + "spawning a request handler thread for queue %d failed", + (int)num); + ret = pthread_setspecific(req->prog->req_queue_key, 0); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, + "resetting request queue in TLS failed"); + } + + rpcsvc_toggle_queue_status( + req->prog, &req->prog->request_queue[num - 1], + req->prog->request_queue_status); + + goto noqueue; + } + } + + pthread_mutex_lock(&queue->queue_lock); + { + empty = list_empty(&queue->request_queue); + + list_add_tail(&req->request_list, &queue->request_queue); + + if (empty && queue->waiting) + pthread_cond_signal(&queue->queue_cond); + } + pthread_mutex_unlock(&queue->queue_lock); + + ret = 0; + } else { + noqueue: + ret = actor_fn(req); + } + } + +err_reply: + + ret = rpcsvc_check_and_reply_error(ret, NULL, req); + /* No need to propagate error beyond this function since the reply + * has now been queued. */ + ret = 0; + +out: + return ret; +} + +int +rpcsvc_handle_disconnect(rpcsvc_t *svc, rpc_transport_t *trans) +{ + rpcsvc_event_t event; + rpcsvc_notify_wrapper_t *wrappers = NULL, *wrapper; + int32_t ret = -1, i = 0, wrapper_count = 0; + rpcsvc_listener_t *listener = NULL; + + event = (trans->listener == NULL) ? RPCSVC_EVENT_LISTENER_DEAD + : RPCSVC_EVENT_DISCONNECT; + + pthread_rwlock_rdlock(&svc->rpclock); + { + if (!svc->notify_count) + goto unlock; + + wrappers = GF_CALLOC(svc->notify_count, sizeof(*wrapper), + gf_common_mt_rpcsvc_wrapper_t); + if (!wrappers) { + goto unlock; + } + + list_for_each_entry(wrapper, &svc->notify, list) + { + if (wrapper->notify) { + wrappers[i++] = *wrapper; + } + } + + wrapper_count = i; + } +unlock: + pthread_rwlock_unlock(&svc->rpclock); + + if (wrappers) { + for (i = 0; i < wrapper_count; i++) { + wrappers[i].notify(svc, wrappers[i].data, event, trans); + } + + GF_FREE(wrappers); + } + + if (event == RPCSVC_EVENT_LISTENER_DEAD) { + listener = rpcsvc_get_listener(svc, -1, trans->listener); + rpcsvc_listener_destroy(listener); + } + + return ret; +} + +int +rpcsvc_notify(rpc_transport_t *trans, void *mydata, rpc_transport_event_t event, + void *data, ...) +{ + int ret = -1; + rpc_transport_pollin_t *msg = NULL; + rpc_transport_t *new_trans = NULL; + rpcsvc_t *svc = NULL; + rpcsvc_listener_t *listener = NULL; + + svc = mydata; + if (svc == NULL) { + goto out; + } + + switch (event) { + case RPC_TRANSPORT_ACCEPT: + new_trans = data; + ret = rpcsvc_accept(svc, trans, new_trans); + break; + + case RPC_TRANSPORT_DISCONNECT: + ret = rpcsvc_handle_disconnect(svc, trans); + break; + + case RPC_TRANSPORT_MSG_RECEIVED: + msg = data; + ret = rpcsvc_handle_rpc_call(svc, trans, msg); + break; + + case RPC_TRANSPORT_MSG_SENT: + ret = 0; + break; + + case RPC_TRANSPORT_CONNECT: + /* do nothing, no need for rpcsvc to handle this, client should + * handle this event + */ + /* print info about transport too : LOG TODO */ + gf_log("rpcsvc", GF_LOG_CRITICAL, + "got CONNECT event, which should have not come"); + ret = 0; + break; + + case RPC_TRANSPORT_CLEANUP: + listener = rpcsvc_get_listener(svc, -1, trans->listener); + if (listener == NULL) { + goto out; + } + + rpcsvc_program_notify(listener, RPCSVC_EVENT_TRANSPORT_DESTROY, + trans); + ret = 0; + break; + + case RPC_TRANSPORT_MAP_XID_REQUEST: + /* FIXME: think about this later */ + gf_log("rpcsvc", GF_LOG_CRITICAL, + "got MAP_XID event, which should have not come"); + ret = 0; + break; + + case RPC_TRANSPORT_EVENT_THREAD_DIED: + rpcsvc_handle_event_thread_death(svc, trans, + (int)(unsigned long)data); + ret = 0; + break; + } + +out: + return ret; +} + +/* Given the RPC reply structure and the payload handed by the RPC program, + * encode the RPC record header into the buffer pointed by recordstart. + */ +struct iovec +rpcsvc_record_build_header(char *recordstart, size_t rlen, struct rpc_msg reply, + size_t payload) +{ + struct iovec replyhdr; + struct iovec txrecord = {0, 0}; + size_t fraglen = 0; + int ret = -1; + + /* After leaving aside the 4 bytes for the fragment header, lets + * encode the RPC reply structure into the buffer given to us. + */ + ret = rpc_reply_to_xdr(&reply, recordstart, rlen, &replyhdr); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "Failed to create RPC reply"); + goto err; + } + + fraglen = payload + replyhdr.iov_len; + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "Reply fraglen %zu, payload: %zu, " + "rpc hdr: %zu", + fraglen, payload, replyhdr.iov_len); + + txrecord.iov_base = recordstart; + + /* Remember, this is only the vec for the RPC header and does not + * include the payload above. We needed the payload only to calculate + * the size of the full fragment. This size is sent in the fragment + * header. + */ + txrecord.iov_len = replyhdr.iov_len; +err: + return txrecord; +} + +static uint32_t +rpc_callback_new_callid(struct rpc_transport *trans) +{ + uint32_t callid = 0; + + pthread_mutex_lock(&trans->lock); + { + callid = ++trans->xid; + } + pthread_mutex_unlock(&trans->lock); + + return callid; +} + +int +rpcsvc_fill_callback(int prognum, int progver, int procnum, int payload, + uint32_t xid, struct rpc_msg *request) +{ + int ret = -1; + + if (!request) { + goto out; + } + + memset(request, 0, sizeof(*request)); + + request->rm_xid = xid; + request->rm_direction = CALL; + + request->rm_call.cb_rpcvers = 2; + request->rm_call.cb_prog = prognum; + request->rm_call.cb_vers = progver; + request->rm_call.cb_proc = procnum; + + request->rm_call.cb_cred.oa_flavor = AUTH_NONE; + request->rm_call.cb_cred.oa_base = NULL; + request->rm_call.cb_cred.oa_length = 0; + + request->rm_call.cb_verf.oa_flavor = AUTH_NONE; + request->rm_call.cb_verf.oa_base = NULL; + request->rm_call.cb_verf.oa_length = 0; + + ret = 0; +out: + return ret; +} + +struct iovec +rpcsvc_callback_build_header(char *recordstart, size_t rlen, + struct rpc_msg *request, size_t payload) +{ + struct iovec requesthdr = { + 0, + }; + struct iovec txrecord = {0, 0}; + int ret = -1; + size_t fraglen = 0; + + ret = rpc_request_to_xdr(request, recordstart, rlen, &requesthdr); + if (ret == -1) { + gf_log("rpcsvc", GF_LOG_WARNING, "Failed to create RPC request"); + goto out; + } + + fraglen = payload + requesthdr.iov_len; + gf_log("rpcsvc", GF_LOG_TRACE, + "Request fraglen %zu, payload: %zu, " + "rpc hdr: %zu", + fraglen, payload, requesthdr.iov_len); + + txrecord.iov_base = recordstart; + + /* Remember, this is only the vec for the RPC header and does not + * include the payload above. We needed the payload only to calculate + * the size of the full fragment. This size is sent in the fragment + * header. + */ + txrecord.iov_len = requesthdr.iov_len; + +out: + return txrecord; +} + +static struct iobuf * +rpcsvc_callback_build_record(rpcsvc_t *rpc, int prognum, int progver, + int procnum, size_t payload, u_long xid, + struct iovec *recbuf) +{ + struct rpc_msg request = { + 0, + }; + struct iobuf *request_iob = NULL; + char *record = NULL; + struct iovec recordhdr = { + 0, + }; + size_t pagesize = 0; + size_t xdr_size = 0; + int ret = -1; + + if ((!rpc) || (!recbuf)) { + goto out; + } + + /* Fill the rpc structure and XDR it into the buffer got above. */ + ret = rpcsvc_fill_callback(prognum, progver, procnum, payload, xid, + &request); + if (ret == -1) { + gf_log("rpcsvc", GF_LOG_WARNING, + "cannot build a rpc-request " + "xid (%lu)", + xid); + goto out; + } + + /* First, try to get a pointer into the buffer which the RPC + * layer can use. + */ + xdr_size = xdr_sizeof((xdrproc_t)xdr_callmsg, &request); + + request_iob = iobuf_get2(rpc->ctx->iobuf_pool, (xdr_size + payload)); + if (!request_iob) { + goto out; + } + + pagesize = iobuf_pagesize(request_iob); + + record = iobuf_ptr(request_iob); /* Now we have it. */ + + recordhdr = rpcsvc_callback_build_header(record, pagesize, &request, + payload); + + if (!recordhdr.iov_base) { + gf_log("rpc-clnt", GF_LOG_ERROR, + "Failed to build record " + " header"); + iobuf_unref(request_iob); + request_iob = NULL; + recbuf->iov_base = NULL; + goto out; + } + + recbuf->iov_base = recordhdr.iov_base; + recbuf->iov_len = recordhdr.iov_len; + +out: + return request_iob; +} + +int +rpcsvc_request_submit(rpcsvc_t *rpc, rpc_transport_t *trans, + rpcsvc_cbk_program_t *prog, int procnum, void *req, + glusterfs_ctx_t *ctx, xdrproc_t xdrproc) +{ + int ret = -1; + int count = 0; + struct iovec iov = { + 0, + }; + struct iobuf *iobuf = NULL; + ssize_t xdr_size = 0; + struct iobref *iobref = NULL; + + if (!req) + goto out; + + xdr_size = xdr_sizeof(xdrproc, req); + + iobuf = iobuf_get2(ctx->iobuf_pool, xdr_size); + if (!iobuf) + goto out; + + iov.iov_base = iobuf->ptr; + iov.iov_len = iobuf_pagesize(iobuf); + + ret = xdr_serialize_generic(iov, req, xdrproc); + if (ret == -1) { + gf_log(THIS->name, GF_LOG_WARNING, "failed to create XDR payload"); + goto out; + } + iov.iov_len = ret; + count = 1; + + iobref = iobref_new(); + if (!iobref) { + ret = -1; + gf_log("rpcsvc", GF_LOG_WARNING, "Failed to create iobref"); + goto out; + } + + iobref_add(iobref, iobuf); + + ret = rpcsvc_callback_submit(rpc, trans, prog, procnum, &iov, count, + iobref); + +out: + if (iobuf) + iobuf_unref(iobuf); + + if (iobref) + iobref_unref(iobref); + + return ret; +} + +int +rpcsvc_callback_submit(rpcsvc_t *rpc, rpc_transport_t *trans, + rpcsvc_cbk_program_t *prog, int procnum, + struct iovec *proghdr, int proghdrcount, + struct iobref *iobref) +{ + struct iobuf *request_iob = NULL; + struct iovec rpchdr = { + 0, + }; + rpc_transport_req_t req; + int ret = -1; + int proglen = 0; + uint32_t xid = 0; + gf_boolean_t new_iobref = _gf_false; + + if (!rpc) { + goto out; + } + + memset(&req, 0, sizeof(req)); + + if (proghdr) { + proglen += iov_length(proghdr, proghdrcount); + } + + xid = rpc_callback_new_callid(trans); + + request_iob = rpcsvc_callback_build_record( + rpc, prog->prognum, prog->progver, procnum, proglen, xid, &rpchdr); + if (!request_iob) { + gf_log("rpcsvc", GF_LOG_WARNING, "cannot build rpc-record"); + goto out; + } + if (!iobref) { + iobref = iobref_new(); + if (!iobref) { + gf_log("rpcsvc", GF_LOG_WARNING, "Failed to create iobref"); + goto out; + } + new_iobref = 1; + } + + iobref_add(iobref, request_iob); + + req.msg.rpchdr = &rpchdr; + req.msg.rpchdrcount = 1; + req.msg.proghdr = proghdr; + req.msg.proghdrcount = proghdrcount; + req.msg.iobref = iobref; + + ret = rpc_transport_submit_request(trans, &req); + if (ret == -1) { + gf_log("rpcsvc", GF_LOG_WARNING, "transmission of rpc-request failed"); + goto out; + } + + ret = 0; + +out: + iobuf_unref(request_iob); + + if (new_iobref) + iobref_unref(iobref); + + return ret; +} + +int +rpcsvc_transport_submit(rpc_transport_t *trans, struct iovec *rpchdr, + int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *progpayload, + int progpayloadcount, struct iobref *iobref, void *priv) +{ + int ret = -1; + rpc_transport_reply_t reply = { + 0, + }; + + if ((!trans) || (!rpchdr) || (!rpchdr->iov_base)) { + goto out; + } + + reply.msg.rpchdr = rpchdr; + reply.msg.rpchdrcount = rpchdrcount; + reply.msg.proghdr = proghdr; + reply.msg.proghdrcount = proghdrcount; + reply.msg.progpayload = progpayload; + reply.msg.progpayloadcount = progpayloadcount; + reply.msg.iobref = iobref; + reply.private = priv; + + ret = rpc_transport_submit_reply(trans, &reply); + +out: + return ret; +} + +int +rpcsvc_fill_reply(rpcsvc_request_t *req, struct rpc_msg *reply) +{ + int ret = -1; + rpcsvc_program_t *prog = NULL; + if ((!req) || (!reply)) + goto out; + + ret = 0; + rpc_fill_empty_reply(reply, req->xid); + if (req->rpc_status == MSG_DENIED) { + rpc_fill_denied_reply(reply, req->rpc_err, req->auth_err); + goto out; + } + + prog = rpcsvc_request_program(req); + + if (req->rpc_status == MSG_ACCEPTED) + rpc_fill_accepted_reply( + reply, req->rpc_err, (prog) ? prog->proglowvers : 0, + (prog) ? prog->proghighvers : 0, req->verf.flavour, + req->verf.datalen, req->verf.authdata); + else + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Invalid rpc_status value"); + +out: + return ret; +} + +/* Given a request and the reply payload, build a reply and encodes the reply + * into a record header. This record header is encoded into the vector pointed + * to be recbuf. + * msgvec is the buffer that points to the payload of the RPC program. + * This buffer can be NULL, if an RPC error reply is being constructed. + * The only reason it is needed here is that in case the buffer is provided, + * we should account for the length of that buffer in the RPC fragment header. + */ +struct iobuf * +rpcsvc_record_build_record(rpcsvc_request_t *req, size_t payload, size_t hdrlen, + struct iovec *recbuf) +{ + struct rpc_msg reply; + struct iobuf *replyiob = NULL; + char *record = NULL; + struct iovec recordhdr = { + 0, + }; + size_t pagesize = 0; + size_t xdr_size = 0; + rpcsvc_t *svc = NULL; + int ret = -1; + + if ((!req) || (!req->trans) || (!req->svc) || (!recbuf)) + return NULL; + + svc = req->svc; + + /* Fill the rpc structure and XDR it into the buffer got above. */ + ret = rpcsvc_fill_reply(req, &reply); + if (ret) + goto err_exit; + + xdr_size = xdr_sizeof((xdrproc_t)xdr_replymsg, &reply); + + /* Payload would include 'readv' size etc too, where as + that comes as another payload iobuf */ + replyiob = iobuf_get2(svc->ctx->iobuf_pool, (xdr_size + hdrlen)); + if (!replyiob) { + goto err_exit; + } + + pagesize = iobuf_pagesize(replyiob); + + record = iobuf_ptr(replyiob); /* Now we have it. */ + + recordhdr = rpcsvc_record_build_header(record, pagesize, reply, payload); + if (!recordhdr.iov_base) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to build record " + " header"); + iobuf_unref(replyiob); + replyiob = NULL; + recbuf->iov_base = NULL; + goto err_exit; + } + + recbuf->iov_base = recordhdr.iov_base; + recbuf->iov_len = recordhdr.iov_len; +err_exit: + return replyiob; +} + +/* + * The function to submit a program message to the RPC service. + * This message is added to the transmission queue of the + * conn. + * + * Program callers are not expected to use the msgvec->iov_base + * address for anything else. + * Nor are they expected to free it once this function returns. + * Once the transmission of the buffer is completed by the RPC service, + * the memory area as referenced through @msg will be unrefed. + * If a higher layer does not want anything to do with this iobuf + * after this function returns, it should call unref on it. For keeping + * it around till the transmission is actually complete, rpcsvc also refs it. + * * + * If this function returns an error by returning -1, the + * higher layer programs should assume that a disconnection happened + * and should know that the conn memory area as well as the req structure + * has been freed internally. + * + * For now, this function assumes that a submit is always called + * to send a new record. Later, if there is a situation where different + * buffers for the same record come from different sources, then we'll + * need to change this code to account for multiple submit calls adding + * the buffers into a single record. + */ + +int +rpcsvc_submit_generic(rpcsvc_request_t *req, struct iovec *proghdr, + int hdrcount, struct iovec *payload, int payloadcount, + struct iobref *iobref) +{ + int ret = -1, i = 0; + struct iobuf *replyiob = NULL; + struct iovec recordhdr = { + 0, + }; + rpc_transport_t *trans = NULL; + size_t msglen = 0; + size_t hdrlen = 0; + char new_iobref = 0; + rpcsvc_drc_globals_t *drc = NULL; + gf_latency_t *lat = NULL; + + if ((!req) || (!req->trans)) + return -1; + + if (req->prog && req->begin.tv_sec) { + if ((req->procnum >= 0) && (req->procnum < req->prog->numactors)) { + timespec_now(&req->end); + lat = &req->prog->latencies[req->procnum]; + gf_latency_update(lat, &req->begin, &req->end); + } + } + trans = req->trans; + + for (i = 0; i < hdrcount; i++) { + msglen += proghdr[i].iov_len; + } + + for (i = 0; i < payloadcount; i++) { + msglen += payload[i].iov_len; + } + + gf_log(GF_RPCSVC, GF_LOG_TRACE, "Tx message: %zu", msglen); + + /* Build the buffer containing the encoded RPC reply. */ + replyiob = rpcsvc_record_build_record(req, msglen, hdrlen, &recordhdr); + if (!replyiob) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Reply record creation failed"); + goto disconnect_exit; + } + + if (!iobref) { + iobref = iobref_new(); + if (!iobref) { + goto disconnect_exit; + } + + new_iobref = 1; + } + + iobref_add(iobref, replyiob); + + /* cache the request in the duplicate request cache for appropriate ops */ + if ((req->reply) && (rpcsvc_need_drc(req))) { + drc = req->svc->drc; + + LOCK(&drc->lock); + ret = rpcsvc_cache_reply(req, iobref, &recordhdr, 1, proghdr, hdrcount, + payload, payloadcount); + UNLOCK(&drc->lock); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "failed to cache reply"); + } + } + + ret = rpcsvc_transport_submit(trans, &recordhdr, 1, proghdr, hdrcount, + payload, payloadcount, iobref, + req->trans_private); + + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "failed to submit message " + "(XID: 0x%x, Program: %s, ProgVers: %d, Proc: %d) to " + "rpc-transport (%s)", + req->xid, req->prog ? req->prog->progname : "(not matched)", + req->prog ? req->prog->progver : 0, req->procnum, + trans ? trans->name : ""); + } else { + gf_log(GF_RPCSVC, GF_LOG_TRACE, + "submitted reply for rpc-message (XID: 0x%x, " + "Program: %s, ProgVers: %d, Proc: %d) to rpc-transport " + "(%s)", + req->xid, req->prog ? req->prog->progname : "-", + req->prog ? req->prog->progver : 0, req->procnum, + trans ? trans->name : ""); + } + +disconnect_exit: + if (replyiob) { + iobuf_unref(replyiob); + } + + if (new_iobref) { + iobref_unref(iobref); + } + + rpcsvc_request_destroy(req); + + return ret; +} + +int +rpcsvc_error_reply(rpcsvc_request_t *req) +{ + struct iovec dummyvec = { + 0, + }; + + if (!req) + return -1; + + gf_log_callingfn("", GF_LOG_DEBUG, "sending a RPC error reply"); + + /* At this point the req should already have been filled with the + * appropriate RPC error numbers. + */ + return rpcsvc_submit_generic(req, &dummyvec, 0, NULL, 0, NULL); +} + +#ifdef IPV6_DEFAULT +int +rpcsvc_program_register_rpcbind6(rpcsvc_program_t *newprog, uint32_t port) +{ + const int IP_BUF_LEN = 64; + char addr_buf[IP_BUF_LEN]; + + int err = 0; + bool_t success = 0; + struct netconfig *nc; + struct netbuf *nb; + + if (!newprog) { + goto out; + } + + nc = getnetconfigent("tcp6"); + if (!nc) { + err = -1; + goto out; + } + + err = sprintf(addr_buf, "::.%d.%d", port >> 8 & 0xff, port & 0xff); + if (err < 0) { + err = -1; + goto out; + } + + nb = uaddr2taddr(nc, addr_buf); + if (!nb) { + err = -1; + goto out; + } + + /* Force the unregistration of the program first. + * This call may fail if nothing has been registered, + * which is fine. + */ + rpcsvc_program_unregister_rpcbind6(newprog); + + success = rpcb_set(newprog->prognum, newprog->progver, nc, nb); + if (!success) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Could not register the IPv6" + " service with rpcbind"); + } + + err = 0; + +out: + return err; +} + +int +rpcsvc_program_unregister_rpcbind6(rpcsvc_program_t *newprog) +{ + int err = 0; + bool_t success = 0; + struct netconfig *nc; + + if (!newprog) { + goto out; + } + + nc = getnetconfigent("tcp6"); + if (!nc) { + err = -1; + goto out; + } + + success = rpcb_unset(newprog->prognum, newprog->progver, nc); + if (!success) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Could not unregister the IPv6" + " service with rpcbind"); + } + + err = 0; +out: + return err; +} +#endif + +/* Register the program with the local portmapper service. */ +int +rpcsvc_program_register_portmap(rpcsvc_program_t *newprog, uint32_t port) +{ + int ret = -1; /* FAIL */ + + if (!newprog) { + goto out; + } + + /* pmap_set() returns 0 for FAIL and 1 for SUCCESS */ + if (!(pmap_set(newprog->prognum, newprog->progver, IPPROTO_TCP, port))) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Could not register with" + " portmap %d %d %u", + newprog->prognum, newprog->progver, port); + goto out; + } + + ret = 0; /* SUCCESS */ +out: + return ret; +} + +int +rpcsvc_program_unregister_portmap(rpcsvc_program_t *prog) +{ + int ret = -1; + + if (!prog) + goto out; + + if (!(pmap_unset(prog->prognum, prog->progver))) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Could not unregister with" + " portmap"); + goto out; + } + + ret = 0; +out: + return ret; +} + +int +rpcsvc_register_portmap_enabled(rpcsvc_t *svc) +{ + return svc->register_portmap; +} + +int32_t +rpcsvc_get_listener_port(rpcsvc_listener_t *listener) +{ + int32_t listener_port = -1; + + if ((listener == NULL) || (listener->trans == NULL)) { + goto out; + } + + switch (listener->trans->myinfo.sockaddr.ss_family) { + case AF_INET: + listener_port = ((struct sockaddr_in *)&listener->trans->myinfo + .sockaddr) + ->sin_port; + break; + + case AF_INET6: + listener_port = ((struct sockaddr_in6 *)&listener->trans->myinfo + .sockaddr) + ->sin6_port; + break; + + default: + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "invalid address family (%d)", + listener->trans->myinfo.sockaddr.ss_family); + goto out; + } + + listener_port = ntohs(listener_port); + +out: + return listener_port; +} + +rpcsvc_listener_t * +rpcsvc_get_listener(rpcsvc_t *svc, uint16_t port, rpc_transport_t *trans) +{ + rpcsvc_listener_t *listener = NULL; + char found = 0; + rpcsvc_listener_t *next = NULL; + uint32_t listener_port = 0; + + if (!svc) { + goto out; + } + + pthread_rwlock_rdlock(&svc->rpclock); + { + list_for_each_entry_safe(listener, next, &svc->listeners, list) + { + if (trans != NULL) { + if (listener->trans == trans) { + found = 1; + break; + } + + continue; + } + + listener_port = rpcsvc_get_listener_port(listener); + if (listener_port == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "invalid port for listener %s", + listener->trans->name); + continue; + } + + if (listener_port == port) { + found = 1; + break; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + + if (!found) { + listener = NULL; + } + +out: + return listener; +} + +/* The only difference between the generic submit and this one is that the + * generic submit is also used for submitting RPC error replies in where there + * are no payloads so the msgvec and msgbuf can be NULL. + * Since RPC programs should be using this function along with their payloads + * we must perform NULL checks before calling the generic submit. + */ +int +rpcsvc_submit_message(rpcsvc_request_t *req, struct iovec *proghdr, + int hdrcount, struct iovec *payload, int payloadcount, + struct iobref *iobref) +{ + if ((!req) || (!req->trans) || (!proghdr) || (!proghdr->iov_base)) + return -1; + + return rpcsvc_submit_generic(req, proghdr, hdrcount, payload, payloadcount, + iobref); +} + +void +rpcsvc_program_destroy(rpcsvc_program_t *program) +{ + if (program) { + GF_FREE(program->latencies); + GF_FREE(program); + } +} + +int +rpcsvc_program_unregister(rpcsvc_t *svc, rpcsvc_program_t *program) +{ + int ret = -1; + rpcsvc_program_t *prog = NULL; + if (!svc || !program) { + goto out; + } + + pthread_rwlock_rdlock(&svc->rpclock); + { + list_for_each_entry(prog, &svc->programs, program) + { + if ((prog->prognum == program->prognum) && + (prog->progver == program->progver)) { + break; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + + ret = rpcsvc_program_unregister_portmap(program); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "portmap unregistration of" + " program failed"); + goto out; + } +#ifdef IPV6_DEFAULT + ret = rpcsvc_program_unregister_rpcbind6(program); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "rpcbind (ipv6)" + " unregistration of program failed"); + goto out; + } +#endif + + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "Program unregistered: %s, Num: %d," + " Ver: %d, Port: %d", + prog->progname, prog->prognum, prog->progver, prog->progport); + + if (prog->ownthread) { + prog->alive = _gf_false; + ret = 0; + goto out; + } + + pthread_rwlock_wrlock(&svc->rpclock); + { + list_del_init(&prog->program); + } + pthread_rwlock_unlock(&svc->rpclock); + + ret = 0; +out: + rpcsvc_program_destroy(prog); + + if (ret == -1) { + if (program) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Program " + "unregistration failed" + ": %s, Num: %d, Ver: %d, Port: %d", + program->progname, program->prognum, program->progver, + program->progport); + } else { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Program not found"); + } + } + + return ret; +} + +int +rpcsvc_transport_peername(rpc_transport_t *trans, char *hostname, int hostlen) +{ + if (!trans) { + return -1; + } + + return rpc_transport_get_peername(trans, hostname, hostlen); +} + +int +rpcsvc_transport_peeraddr(rpc_transport_t *trans, char *addrstr, int addrlen, + struct sockaddr_storage *sa, socklen_t sasize) +{ + if (!trans) { + return -1; + } + + return rpc_transport_get_peeraddr(trans, addrstr, addrlen, sa, sasize); +} + +rpcsvc_listener_t * +rpcsvc_listener_alloc(rpcsvc_t *svc, rpc_transport_t *trans) +{ + rpcsvc_listener_t *listener = NULL; + + listener = GF_CALLOC(1, sizeof(*listener), gf_common_mt_rpcsvc_listener_t); + if (!listener) { + goto out; + } + + listener->trans = trans; + listener->svc = svc; + + INIT_LIST_HEAD(&listener->list); + + pthread_rwlock_wrlock(&svc->rpclock); + { + list_add_tail(&listener->list, &svc->listeners); + } + pthread_rwlock_unlock(&svc->rpclock); +out: + return listener; +} + +int32_t +rpcsvc_create_listener(rpcsvc_t *svc, dict_t *options, char *name) +{ + rpc_transport_t *trans = NULL; + rpcsvc_listener_t *listener = NULL; + int32_t ret = -1; + + if (!svc || !options) { + goto out; + } + + trans = rpc_transport_load(svc->ctx, options, name); + if (!trans) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, + "cannot create listener, " + "initing the transport failed"); + goto out; + } + + ret = rpc_transport_listen(trans); + if (ret == -EADDRINUSE || ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "listening on transport failed"); + goto out; + } + + ret = rpc_transport_register_notify(trans, rpcsvc_notify, svc); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_WARNING, "registering notify failed"); + goto out; + } + + listener = rpcsvc_listener_alloc(svc, trans); + if (listener == NULL) { + ret = -1; + goto out; + } + + ret = 0; +out: + if (!listener && trans) { + rpc_transport_disconnect(trans, _gf_true); + rpc_transport_cleanup(trans); + } + + return ret; +} + +int32_t +rpcsvc_create_listeners(rpcsvc_t *svc, dict_t *options, char *name) +{ + int32_t ret = -1, count = 0; + data_t *data = NULL; + char *str = NULL, *ptr = NULL, *transport_name = NULL; + char *transport_type = NULL, *saveptr = NULL, *tmp = NULL; + + if ((svc == NULL) || (options == NULL) || (name == NULL)) { + goto out; + } + + data = dict_get(options, "transport-type"); + if (data == NULL) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "option transport-type not set"); + goto out; + } + + transport_type = data_to_str(data); + if (transport_type == NULL) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "option transport-type not set"); + goto out; + } + + /* duplicate transport_type, since following dict_set will free it */ + transport_type = gf_strdup(transport_type); + if (transport_type == NULL) { + goto out; + } + + str = gf_strdup(transport_type); + if (str == NULL) { + goto out; + } + + ptr = strtok_r(str, ",", &saveptr); + + while (ptr != NULL) { + tmp = gf_strdup(ptr); + if (tmp == NULL) { + goto out; + } + + ret = gf_asprintf(&transport_name, "%s.%s", tmp, name); + if (ret == -1) { + goto out; + } + + ret = dict_set_dynstr(options, "transport-type", tmp); + if (ret == -1) { + goto out; + } + + tmp = NULL; + ptr = strtok_r(NULL, ",", &saveptr); + + ret = rpcsvc_create_listener(svc, options, transport_name); + if (ret != 0) { + goto out; + } + + dict_del(options, "notify-poller-death"); + GF_FREE(transport_name); + transport_name = NULL; + count++; + } + + ret = dict_set_dynstr(options, "transport-type", transport_type); + if (ret == -1) { + goto out; + } + + transport_type = NULL; + +out: + GF_FREE(str); + + GF_FREE(transport_type); + + GF_FREE(tmp); + + GF_FREE(transport_name); + + if (count > 0) { + return count; + } else { + return ret; + } +} + +int +rpcsvc_unregister_notify(rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata) +{ + rpcsvc_notify_wrapper_t *wrapper = NULL, *tmp = NULL; + int ret = 0; + + if (!svc || !notify) { + goto out; + } + + pthread_rwlock_wrlock(&svc->rpclock); + { + list_for_each_entry_safe(wrapper, tmp, &svc->notify, list) + { + if ((wrapper->notify == notify) && (mydata == wrapper->data)) { + list_del_init(&wrapper->list); + GF_FREE(wrapper); + ret++; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + +out: + return ret; +} + +int +rpcsvc_register_notify(rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata) +{ + rpcsvc_notify_wrapper_t *wrapper = NULL; + int ret = -1; + + wrapper = rpcsvc_notify_wrapper_alloc(); + if (!wrapper) { + goto out; + } + svc->mydata = mydata; + wrapper->data = mydata; + wrapper->notify = notify; + + pthread_rwlock_wrlock(&svc->rpclock); + { + list_add_tail(&wrapper->list, &svc->notify); + svc->notify_count++; + } + pthread_rwlock_unlock(&svc->rpclock); + + ret = 0; +out: + return ret; +} + +void * +rpcsvc_request_handler(void *arg) +{ + rpcsvc_request_queue_t *queue = NULL; + rpcsvc_program_t *program = NULL; + rpcsvc_request_t *req = NULL, *tmp_req = NULL; + rpcsvc_actor_t *actor = NULL; + gf_boolean_t done = _gf_false; + int ret = 0; + struct list_head tmp_list; + + queue = arg; + program = queue->program; + + INIT_LIST_HEAD(&tmp_list); + + if (!program) + return NULL; + + while (1) { + pthread_mutex_lock(&queue->queue_lock); + { + if (!program->alive && list_empty(&queue->request_queue)) { + done = 1; + goto unlock; + } + + while (list_empty(&queue->request_queue)) { + queue->waiting = _gf_true; + pthread_cond_wait(&queue->queue_cond, &queue->queue_lock); + } + + queue->waiting = _gf_false; + + if (!list_empty(&queue->request_queue)) { + INIT_LIST_HEAD(&tmp_list); + list_splice_init(&queue->request_queue, &tmp_list); + } + } + unlock: + pthread_mutex_unlock(&queue->queue_lock); + + list_for_each_entry_safe(req, tmp_req, &tmp_list, request_list) + { + if (req) { + list_del_init(&req->request_list); + + if (req->prognum == RPCSVC_INFRA_PROGRAM) { + switch (req->procnum) { + case RPCSVC_PROC_EVENT_THREAD_DEATH: + gf_log(GF_RPCSVC, GF_LOG_INFO, + "event thread died, exiting request handler " + "thread for queue %d of program %s", + (int)(queue - &program->request_queue[0]), + program->progname); + done = 1; + pthread_mutex_lock(&program->thr_lock); + { + rpcsvc_toggle_queue_status( + program, queue, + program->request_queue_status); + program->threadcount--; + } + pthread_mutex_unlock(&program->thr_lock); + rpcsvc_request_destroy(req); + break; + + default: + break; + } + } else { + THIS = req->svc->xl; + actor = rpcsvc_program_actor(req); + ret = actor->actor(req); + + if (ret != 0) { + rpcsvc_check_and_reply_error(ret, NULL, req); + } + req = NULL; + } + } + } + + if (done) + break; + } + + return NULL; +} + +int +rpcsvc_program_register(rpcsvc_t *svc, rpcsvc_program_t *program, + gf_boolean_t add_to_head) +{ + int ret = -1, i = 0; + rpcsvc_program_t *newprog = NULL; + char already_registered = 0; + pthread_mutexattr_t attr[EVENT_MAX_THREADS]; + pthread_mutexattr_t thr_attr; + + if (!svc) { + goto out; + } + + if (program->actors == NULL) { + goto out; + } + + pthread_rwlock_rdlock(&svc->rpclock); + { + list_for_each_entry(newprog, &svc->programs, program) + { + if ((newprog->prognum == program->prognum) && + (newprog->progver == program->progver)) { + already_registered = 1; + break; + } + } + } + pthread_rwlock_unlock(&svc->rpclock); + + if (already_registered) { + ret = 0; + goto out; + } + + newprog = GF_CALLOC(1, sizeof(*newprog), gf_common_mt_rpcsvc_program_t); + if (newprog == NULL) { + goto out; + } + + memcpy(newprog, program, sizeof(*program)); + newprog->latencies = gf_latency_new(program->numactors); + if (!newprog->latencies) { + rpcsvc_program_destroy(newprog); + goto out; + } + + INIT_LIST_HEAD(&newprog->program); + pthread_mutexattr_init(&thr_attr); + pthread_mutexattr_settype(&thr_attr, PTHREAD_MUTEX_ADAPTIVE_NP); + + for (i = 0; i < EVENT_MAX_THREADS; i++) { + pthread_mutexattr_init(&attr[i]); + pthread_mutexattr_settype(&attr[i], PTHREAD_MUTEX_ADAPTIVE_NP); + INIT_LIST_HEAD(&newprog->request_queue[i].request_queue); + pthread_mutex_init(&newprog->request_queue[i].queue_lock, &attr[i]); + pthread_cond_init(&newprog->request_queue[i].queue_cond, NULL); + newprog->request_queue[i].program = newprog; + } + + pthread_mutex_init(&newprog->thr_lock, &thr_attr); + pthread_cond_init(&newprog->thr_cond, NULL); + + newprog->alive = _gf_true; + + if (gf_async_ctrl.enabled) { + newprog->ownthread = _gf_false; + newprog->synctask = _gf_false; + } + + /* make sure synctask gets priority over ownthread */ + if (newprog->synctask) + newprog->ownthread = _gf_false; + + if (newprog->ownthread) { + struct event_pool *ep = svc->ctx->event_pool; + newprog->eventthreadcount = ep->eventthreadcount; + + pthread_key_create(&newprog->req_queue_key, NULL); + newprog->thr_queue = 1; + } + + pthread_rwlock_wrlock(&svc->rpclock); + { + if (add_to_head) + list_add(&newprog->program, &svc->programs); + else + list_add_tail(&newprog->program, &svc->programs); + } + pthread_rwlock_unlock(&svc->rpclock); + + ret = 0; + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "New program registered: %s, Num: %d," + " Ver: %d, Port: %d", + newprog->progname, newprog->prognum, newprog->progver, + newprog->progport); + +out: + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Program registration failed:" + " %s, Num: %d, Ver: %d, Port: %d", + program->progname, program->prognum, program->progver, + program->progport); + } + + return ret; +} + +static void +free_prog_details(gf_dump_rsp *rsp) +{ + gf_prog_detail *prev = NULL; + gf_prog_detail *trav = NULL; + + trav = rsp->prog; + while (trav) { + prev = trav; + trav = trav->next; + GF_FREE(prev); + } +} + +static int +build_prog_details(rpcsvc_request_t *req, gf_dump_rsp *rsp) +{ + int ret = -1; + rpcsvc_program_t *program = NULL; + gf_prog_detail *prog = NULL; + gf_prog_detail *prev = NULL; + + if (!req || !req->trans || !req->svc) + goto out; + + pthread_rwlock_rdlock(&req->svc->rpclock); + { + list_for_each_entry(program, &req->svc->programs, program) + { + prog = GF_CALLOC(1, sizeof(*prog), 0); + if (!prog) + goto unlock; + + prog->progname = program->progname; + prog->prognum = program->prognum; + prog->progver = program->progver; + + if (!rsp->prog) + rsp->prog = prog; + if (prev) + prev->next = prog; + prev = prog; + } + if (prev) + ret = 0; + } +unlock: + pthread_rwlock_unlock(&req->svc->rpclock); +out: + return ret; +} + +static int +rpcsvc_ping(rpcsvc_request_t *req) +{ + char rsp_buf[8 * 1024] = { + 0, + }; + gf_common_rsp rsp = { + 0, + }; + struct iovec iov = { + 0, + }; + int ret = -1; + uint32_t ping_rsp_len = 0; + + ping_rsp_len = xdr_sizeof((xdrproc_t)xdr_gf_common_rsp, &rsp); + + iov.iov_base = rsp_buf; + iov.iov_len = ping_rsp_len; + + ret = xdr_serialize_generic(iov, &rsp, (xdrproc_t)xdr_gf_common_rsp); + if (ret < 0) { + ret = RPCSVC_ACTOR_ERROR; + } else { + rsp.op_ret = 0; + rpcsvc_submit_generic(req, &iov, 1, NULL, 0, NULL); + } + + return 0; +} + +static int +rpcsvc_dump(rpcsvc_request_t *req) +{ + char rsp_buf[8 * 1024] = { + 0, + }; + gf_dump_rsp rsp = { + 0, + }; + struct iovec iov = { + 0, + }; + int op_errno = EINVAL; + int ret = -1; + uint32_t dump_rsp_len = 0; + + if (!req) + goto sendrsp; + + ret = build_prog_details(req, &rsp); + if (ret < 0) { + op_errno = -ret; + goto sendrsp; + } + + op_errno = 0; + +sendrsp: + rsp.op_errno = gf_errno_to_error(op_errno); + rsp.op_ret = ret; + + dump_rsp_len = xdr_sizeof((xdrproc_t)xdr_gf_dump_rsp, &rsp); + + iov.iov_base = rsp_buf; + iov.iov_len = dump_rsp_len; + + ret = xdr_serialize_generic(iov, &rsp, (xdrproc_t)xdr_gf_dump_rsp); + if (ret < 0) { + ret = RPCSVC_ACTOR_ERROR; + } else { + rpcsvc_submit_generic(req, &iov, 1, NULL, 0, NULL); + ret = 0; + } + + free_prog_details(&rsp); + + return ret; +} + +int +rpcsvc_init_options(rpcsvc_t *svc, dict_t *options) +{ + char *optstr = NULL; + int ret = -1; + + if ((!svc) || (!options)) + return -1; + + svc->memfactor = RPCSVC_DEFAULT_MEMFACTOR; + + svc->register_portmap = _gf_true; + if (dict_get(options, "rpc.register-with-portmap")) { + ret = dict_get_str(options, "rpc.register-with-portmap", &optstr); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to parse " + "dict"); + goto out; + } + + ret = gf_string2boolean(optstr, &svc->register_portmap); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to parse bool " + "string"); + goto out; + } + } + + if (!svc->register_portmap) + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "Portmap registration " + "disabled"); + ret = 0; +out: + return ret; +} + +int +rpcsvc_reconfigure_options(rpcsvc_t *svc, dict_t *options) +{ + xlator_t *xlator = NULL; + xlator_list_t *volentry = NULL; + char *srchkey = NULL; + char *keyval = NULL; + int ret = -1; + + if ((!svc) || (!svc->options) || (!options)) + return (-1); + + /* Fetch the xlator from svc */ + xlator = svc->xl; + if (!xlator) + return (-1); + + /* Reconfigure the volume specific rpc-auth.addr allow part */ + volentry = xlator->children; + while (volentry) { + ret = gf_asprintf(&srchkey, "rpc-auth.addr.%s.allow", + volentry->xlator->name); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return (-1); + } + + /* key-string: rpc-auth.addr.<volname>.allow + * + * IMP: Delete the OLD key/value pair from dict. + * And set the NEW key/value pair IFF the option is SET + * in reconfigured volfile. + * + * NB: If rpc-auth.addr.<volname>.allow is not SET explicitly, + * build_nfs_graph() sets it as "*" i.e. anonymous. + */ + dict_del(svc->options, srchkey); + if (!dict_get_str(options, srchkey, &keyval)) { + ret = dict_set_dynstr_with_alloc(svc->options, srchkey, keyval); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "dict_set_str error"); + GF_FREE(srchkey); + return (-1); + } + } + + GF_FREE(srchkey); + volentry = volentry->next; + } + + /* Reconfigure the volume specific rpc-auth.addr reject part */ + volentry = xlator->children; + while (volentry) { + ret = gf_asprintf(&srchkey, "rpc-auth.addr.%s.reject", + volentry->xlator->name); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return (-1); + } + + /* key-string: rpc-auth.addr.<volname>.reject + * + * IMP: Delete the OLD key/value pair from dict. + * And set the NEW key/value pair IFF the option is SET + * in reconfigured volfile. + * + * NB: No default value for reject key. + */ + dict_del(svc->options, srchkey); + if (!dict_get_str(options, srchkey, &keyval)) { + ret = dict_set_dynstr_with_alloc(svc->options, srchkey, keyval); + if (ret < 0) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "dict_set_str error"); + GF_FREE(srchkey); + return (-1); + } + } + + GF_FREE(srchkey); + volentry = volentry->next; + } + + ret = rpcsvc_init_options(svc, options); + if (ret) + return (-1); + + return rpcsvc_auth_reconf(svc, options); +} + +int +rpcsvc_transport_unix_options_build(dict_t *dict, char *filepath) +{ + char *fpath = NULL; + int ret = -1; + + GF_ASSERT(filepath); + GF_VALIDATE_OR_GOTO("rpcsvc", dict, out); + + fpath = gf_strdup(filepath); + if (!fpath) { + ret = -1; + goto out; + } + + ret = dict_set_dynstr(dict, "transport.socket.listen-path", fpath); + if (ret) + goto out; + + ret = dict_set_str(dict, "transport.address-family", "unix"); + if (ret) + goto out; + + ret = dict_set_str(dict, "transport.socket.nodelay", "off"); + if (ret) + goto out; + + ret = dict_set_str(dict, "transport-type", "socket"); + if (ret) + goto out; +out: + if (ret) { + GF_FREE(fpath); + } + return ret; +} + +/* + * Configure() the rpc.outstanding-rpc-limit param. + * If dict_get_int32() for dict-key "rpc.outstanding-rpc-limit" FAILS, + * it would set the value as "defvalue". Otherwise it would fetch the + * value and round up to multiple-of-8. defvalue must be +ve. + * + * NB: defval or set-value "0" is special which means unlimited/65536. + */ +int +rpcsvc_set_outstanding_rpc_limit(rpcsvc_t *svc, dict_t *options, int defvalue) +{ + int ret = -1; /* FAILURE */ + int rpclim = 0; + static char *rpclimkey = "rpc.outstanding-rpc-limit"; + + if ((!svc) || (!options)) + return (-1); + + if ((defvalue < RPCSVC_MIN_OUTSTANDING_RPC_LIMIT) || + (defvalue > RPCSVC_MAX_OUTSTANDING_RPC_LIMIT)) { + return (-1); + } + + /* Fetch the rpc.outstanding-rpc-limit from dict. */ + ret = dict_get_int32(options, rpclimkey, &rpclim); + if (ret < 0) { + /* Fall back to default for FAILURE */ + rpclim = defvalue; + } + + /* Round up to multiple-of-8. It must not exceed + * RPCSVC_MAX_OUTSTANDING_RPC_LIMIT. + */ + rpclim = ((rpclim + 8 - 1) >> 3) * 8; + if (rpclim > RPCSVC_MAX_OUTSTANDING_RPC_LIMIT) { + rpclim = RPCSVC_MAX_OUTSTANDING_RPC_LIMIT; + } + + if (svc->outstanding_rpc_limit != rpclim) { + svc->outstanding_rpc_limit = rpclim; + gf_log(GF_RPCSVC, GF_LOG_INFO, "Configured %s with value %d", rpclimkey, + rpclim); + } + + return (0); +} + +/* + * Enable throttling for rpcsvc_t svc. + * Returns 0 on success, -1 otherwise. + */ +int +rpcsvc_set_throttle_on(rpcsvc_t *svc) +{ + if (!svc) + return -1; + + svc->throttle = _gf_true; + + return 0; +} + +/* + * Disable throttling for rpcsvc_t svc. + * Returns 0 on success, -1 otherwise. + */ +int +rpcsvc_set_throttle_off(rpcsvc_t *svc) +{ + if (!svc) + return -1; + + svc->throttle = _gf_false; + + return 0; +} + +/* + * Get throttle state for rpcsvc_t svc. + * Returns value of attribute throttle on success, _gf_false otherwise. + */ +gf_boolean_t +rpcsvc_get_throttle(rpcsvc_t *svc) +{ + if (!svc) + return _gf_false; + + return svc->throttle; +} + +/* Function call to cleanup resources for svc + */ +int +rpcsvc_destroy(rpcsvc_t *svc) +{ + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + rpcsvc_listener_t *listener = NULL; + rpcsvc_listener_t *next = NULL; + int ret = 0; + + if (!svc) + return ret; + + list_for_each_entry_safe(listener, next, &svc->listeners, list) + { + rpcsvc_listener_destroy(listener); + } + + list_for_each_entry_safe(auth, tmp, &svc->authschemes, authlist) + { + list_del_init(&auth->authlist); + GF_FREE(auth); + } + + rpcsvc_program_unregister(svc, &gluster_dump_prog); + if (svc->rxpool) { + mem_pool_destroy(svc->rxpool); + svc->rxpool = NULL; + } + + pthread_rwlock_destroy(&svc->rpclock); + GF_FREE(svc); + + return ret; +} + +/* The global RPC service initializer. + */ +rpcsvc_t * +rpcsvc_init(xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options, + uint32_t poolcount) +{ + rpcsvc_t *svc = NULL; + int ret = -1; + + if ((!xl) || (!ctx) || (!options)) + return NULL; + + svc = GF_CALLOC(1, sizeof(*svc), gf_common_mt_rpcsvc_t); + if (!svc) + return NULL; + + pthread_rwlock_init(&svc->rpclock, NULL); + INIT_LIST_HEAD(&svc->authschemes); + INIT_LIST_HEAD(&svc->notify); + INIT_LIST_HEAD(&svc->listeners); + INIT_LIST_HEAD(&svc->programs); + + ret = rpcsvc_init_options(svc, options); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "Failed to init options"); + goto free_svc; + } + + if (!poolcount) + poolcount = RPCSVC_POOLCOUNT_MULT * svc->memfactor; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, "rx pool: %d", poolcount); + svc->rxpool = mem_pool_new(rpcsvc_request_t, poolcount); + /* TODO: leak */ + if (!svc->rxpool) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "mem pool allocation failed"); + goto free_svc; + } + + ret = rpcsvc_auth_init(svc, options); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to init " + "authentication"); + goto free_svc; + } + + ret = -1; + svc->options = options; + svc->ctx = ctx; + svc->xl = xl; + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "RPC service inited."); + + gluster_dump_prog.options = options; + + ret = rpcsvc_program_register(svc, &gluster_dump_prog, _gf_false); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "failed to register DUMP program"); + goto free_svc; + } + + ret = 0; +free_svc: + if (ret == -1) { + GF_FREE(svc); + svc = NULL; + } + + return svc; +} + +int +rpcsvc_transport_peer_check_search(dict_t *options, char *pattern, char *ip, + char *hostname) +{ + int ret = -1; + char *addrtok = NULL; + char *addrstr = NULL; + char *dup_addrstr = NULL; + char *svptr = NULL; + + if ((!options) || (!ip)) + return -1; + + ret = dict_get_str(options, pattern, &addrstr); + if (ret < 0) { + ret = -1; + goto err; + } + + if (!addrstr) { + ret = -1; + goto err; + } + + dup_addrstr = gf_strdup(addrstr); + if (dup_addrstr == NULL) { + ret = -1; + goto err; + } + addrtok = strtok_r(dup_addrstr, ",", &svptr); + while (addrtok) { + /* CASEFOLD not present on Solaris */ +#ifdef FNM_CASEFOLD + ret = fnmatch(addrtok, ip, FNM_CASEFOLD); +#else + ret = fnmatch(addrtok, ip, 0); +#endif + if (ret == 0) + goto err; + + /* compare hostnames if applicable */ + if (hostname) { +#ifdef FNM_CASEFOLD + ret = fnmatch(addrtok, hostname, FNM_CASEFOLD); +#else + ret = fnmatch(addrtok, hostname, 0); +#endif + if (ret == 0) + goto err; + } + + /* Compare IPv4 subnetwork, TODO: IPv6 subnet support */ + if (strchr(addrtok, '/')) { + ret = rpcsvc_match_subnet_v4(addrtok, ip); + if (ret == 0) + goto err; + } + + addrtok = strtok_r(NULL, ",", &svptr); + } + + ret = -1; +err: + GF_FREE(dup_addrstr); + + return ret; +} + +static int +rpcsvc_transport_peer_check_allow(dict_t *options, char *volname, char *ip, + char *hostname) +{ + int ret = RPCSVC_AUTH_DONTCARE; + char *srchstr = NULL; + + if ((!options) || (!ip) || (!volname)) + return ret; + + ret = gf_asprintf(&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_DONTCARE; + goto out; + } + + ret = rpcsvc_transport_peer_check_search(options, srchstr, ip, hostname); + GF_FREE(srchstr); + + if (ret == 0) + ret = RPCSVC_AUTH_ACCEPT; + else + ret = RPCSVC_AUTH_REJECT; +out: + return ret; +} + +static int +rpcsvc_transport_peer_check_reject(dict_t *options, char *volname, char *ip, + char *hostname) +{ + int ret = RPCSVC_AUTH_DONTCARE; + char *srchstr = NULL; + + if ((!options) || (!ip) || (!volname)) + return ret; + + ret = gf_asprintf(&srchstr, "rpc-auth.addr.%s.reject", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_REJECT; + goto out; + } + + ret = rpcsvc_transport_peer_check_search(options, srchstr, ip, hostname); + GF_FREE(srchstr); + + if (ret == 0) + ret = RPCSVC_AUTH_REJECT; + else + ret = RPCSVC_AUTH_DONTCARE; +out: + return ret; +} + +/* Combines rpc auth's allow and reject options. + * Order of checks is important. + * First, REJECT if either rejects. + * If neither rejects, ACCEPT if either accepts. + * If neither accepts, DONTCARE + */ +int +rpcsvc_combine_allow_reject_volume_check(int allow, int reject) +{ + if (allow == RPCSVC_AUTH_REJECT || reject == RPCSVC_AUTH_REJECT) + return RPCSVC_AUTH_REJECT; + + if (allow == RPCSVC_AUTH_ACCEPT || reject == RPCSVC_AUTH_ACCEPT) + return RPCSVC_AUTH_ACCEPT; + + return RPCSVC_AUTH_DONTCARE; +} + +int +rpcsvc_auth_check(rpcsvc_t *svc, char *volname, char *ipaddr) +{ + int ret = RPCSVC_AUTH_REJECT; + int accept = RPCSVC_AUTH_REJECT; + int reject = RPCSVC_AUTH_REJECT; + char *hostname = NULL; + char *allow_str = NULL; + char *reject_str = NULL; + char *srchstr = NULL; + dict_t *options = NULL; + + if (!svc || !volname || !ipaddr) + return ret; + + /* Fetch the options from svc struct and validate */ + options = svc->options; + if (!options) + return ret; + + /* Accept if its the default case: Allow all, Reject none + * The default volfile always contains a 'allow *' rule + * for each volume. If allow rule is missing (which implies + * there is some bad volfile generating code doing this), we + * assume no one is allowed mounts, and thus, we reject mounts. + */ + ret = gf_asprintf(&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return RPCSVC_AUTH_REJECT; + } + + ret = dict_get_str(options, srchstr, &allow_str); + GF_FREE(srchstr); + if (ret < 0) + return RPCSVC_AUTH_REJECT; + + ret = gf_asprintf(&srchstr, "rpc-auth.addr.%s.reject", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return RPCSVC_AUTH_REJECT; + } + + ret = dict_get_str(options, srchstr, &reject_str); + GF_FREE(srchstr); + + /* + * If "reject_str" is being set as '*' (anonymous), then NFS-server + * would reject everything. If the "reject_str" is not set and + * "allow_str" is set as '*' (anonymous), then NFS-server would + * accept mount requests from all clients. + */ + if (reject_str != NULL) { + if (!strcmp("*", reject_str)) + return RPCSVC_AUTH_REJECT; + } else { + if (!strcmp("*", allow_str)) + return RPCSVC_AUTH_ACCEPT; + } + + /* addr-namelookup check */ + if (svc->addr_namelookup == _gf_true) { + ret = gf_get_hostname_from_ip(ipaddr, &hostname); + if (ret) { + if (hostname) + GF_FREE(hostname); + /* failed to get hostname, but hostname auth + * is enabled, so authentication will not be + * 100% correct. reject mounts + */ + return RPCSVC_AUTH_REJECT; + } + } + + accept = rpcsvc_transport_peer_check_allow(options, volname, ipaddr, + hostname); + + reject = rpcsvc_transport_peer_check_reject(options, volname, ipaddr, + hostname); + + if (hostname) + GF_FREE(hostname); + return rpcsvc_combine_allow_reject_volume_check(accept, reject); +} + +int +rpcsvc_transport_privport_check(rpcsvc_t *svc, char *volname, uint16_t port) +{ + int ret = RPCSVC_AUTH_REJECT; + char *srchstr = NULL; + char *valstr = NULL; + gf_boolean_t insecure = _gf_false; + + if ((!svc) || (!volname)) + return ret; + + gf_log(GF_RPCSVC, GF_LOG_TRACE, "Client port: %d", (int)port); + /* If the port is already a privileged one, don't bother with checking + * options. + */ + if (port <= 1024) { + ret = RPCSVC_AUTH_ACCEPT; + goto err; + } + + /* Disabled by default */ + ret = gf_asprintf(&srchstr, "rpc-auth.ports.%s.insecure", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + ret = dict_get_str(svc->options, srchstr, &valstr); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to" + " read rpc-auth.ports.insecure value"); + goto err; + } + + ret = gf_string2boolean(valstr, &insecure); + if (ret) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, + "Failed to" + " convert rpc-auth.ports.insecure value"); + goto err; + } + + ret = insecure ? RPCSVC_AUTH_ACCEPT : RPCSVC_AUTH_REJECT; + + if (ret == RPCSVC_AUTH_ACCEPT) + gf_log(GF_RPCSVC, GF_LOG_DEBUG, "Unprivileged port allowed"); + else + gf_log(GF_RPCSVC, GF_LOG_DEBUG, + "Unprivileged port not" + " allowed"); + +err: + if (srchstr) + GF_FREE(srchstr); + + return ret; +} + +char * +rpcsvc_volume_allowed(dict_t *options, char *volname) +{ + char globalrule[] = "rpc-auth.addr.allow"; + char *srchstr = NULL; + char *addrstr = NULL; + int ret = -1; + + if ((!options) || (!volname)) + return NULL; + + ret = gf_asprintf(&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log(GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + goto out; + } + + if (!dict_get(options, srchstr)) + ret = dict_get_str(options, globalrule, &addrstr); + else + ret = dict_get_str(options, srchstr, &addrstr); + +out: + GF_FREE(srchstr); + + return addrstr; +} + +/* + * rpcsvc_match_subnet_v4() takes subnetwork address pattern and checks + * if the target IPv4 address has the same network address with the help + * of network mask. + * + * Returns 0 for SUCCESS and -1 otherwise. + * + * NB: Validation of subnetwork address pattern is not required + * as it's already being done at the time of CLI SET. + */ +static int +rpcsvc_match_subnet_v4(const char *addrtok, const char *ipaddr) +{ + char *slash = NULL; + char *netaddr = NULL; + int ret = -1; + uint32_t prefixlen = 0; + uint32_t shift = 0; + struct sockaddr_in sin1 = { + 0, + }; + struct sockaddr_in sin2 = { + 0, + }; + struct sockaddr_in mask = { + 0, + }; + + /* Copy the input */ + netaddr = gf_strdup(addrtok); + if (netaddr == NULL) /* ENOMEM */ + goto out; + + /* Find the network socket addr of target */ + if (inet_pton(AF_INET, ipaddr, &sin1.sin_addr) == 0) + goto out; + + slash = strchr(netaddr, '/'); + if (slash) { + *slash = '\0'; + /* + * Find the IPv4 network mask in network byte order. + * IMP: String slash+1 is already validated, it can't have value + * more than IPv4_ADDR_SIZE (32). + */ + prefixlen = (uint32_t)atoi(slash + 1); + if (prefixlen > 31) + goto out; + } else { + /* if there is no '/', then this function wouldn't be called */ + goto out; + } + + /* Need to do this after removing '/', as inet_pton() take IP address as + * second argument. Once we get sin2, then comparison is oranges to orange + */ + if (inet_pton(AF_INET, netaddr, &sin2.sin_addr) == 0) + goto out; + + shift = IPv4_ADDR_SIZE - prefixlen; + mask.sin_addr.s_addr = htonl((uint32_t)~0 << shift); + + if (mask_match(sin1.sin_addr.s_addr, sin2.sin_addr.s_addr, + mask.sin_addr.s_addr)) { + ret = 0; /* SUCCESS */ + } +out: + GF_FREE(netaddr); + return ret; +} + +void +rpcsvc_program_dump(rpcsvc_program_t *prog) +{ + char key_prefix[GF_DUMP_MAX_BUF_LEN]; + char key[GF_DUMP_MAX_BUF_LEN]; + int i; + + snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s", prog->progname); + gf_proc_dump_add_section("%s", key_prefix); + + gf_proc_dump_build_key(key, key_prefix, "program-number"); + gf_proc_dump_write(key, "%d", prog->prognum); + + gf_proc_dump_build_key(key, key_prefix, "program-version"); + gf_proc_dump_write(key, "%d", prog->progver); + + strncat(key_prefix, ".latency", + sizeof(key_prefix) - strlen(key_prefix) - 1); + + for (i = 0; i < prog->numactors; i++) { + gf_proc_dump_build_key(key, key_prefix, "%s", prog->actors[i].procname); + gf_latency_statedump_and_reset(key, &prog->latencies[i]); + } +} + +void +rpcsvc_statedump(rpcsvc_t *svc) +{ + rpcsvc_program_t *prog = NULL; + int ret = 0; + ret = pthread_rwlock_tryrdlock(&svc->rpclock); + if (ret) + return; + { + list_for_each_entry(prog, &svc->programs, program) + { + rpcsvc_program_dump(prog); + } + } + pthread_rwlock_unlock(&svc->rpclock); +} + +static rpcsvc_actor_t gluster_dump_actors[GF_DUMP_MAXVALUE] = { + [GF_DUMP_NULL] = {"NULL", NULL, NULL, GF_DUMP_NULL, DRC_NA, 0}, + [GF_DUMP_DUMP] = {"DUMP", rpcsvc_dump, NULL, GF_DUMP_DUMP, DRC_NA, 0}, + [GF_DUMP_PING] = {"PING", rpcsvc_ping, NULL, GF_DUMP_PING, DRC_NA, 0}, +}; + +static struct rpcsvc_program gluster_dump_prog = { + .progname = "GF-DUMP", + .prognum = GLUSTER_DUMP_PROGRAM, + .progver = GLUSTER_DUMP_VERSION, + .actors = gluster_dump_actors, + .numactors = GF_DUMP_MAXVALUE, +}; diff --git a/rpc/rpc-lib/src/rpcsvc.h b/rpc/rpc-lib/src/rpcsvc.h new file mode 100644 index 00000000000..7b3030926c8 --- /dev/null +++ b/rpc/rpc-lib/src/rpcsvc.h @@ -0,0 +1,694 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _RPCSVC_H +#define _RPCSVC_H + +#include <glusterfs/gf-event.h> +#include "rpc-transport.h" +#include <glusterfs/dict.h> +#include "xdr-rpc.h" +#include "rpcsvc-common.h" + +#include <pthread.h> +#include <sys/uio.h> +#include <inttypes.h> +#include <rpc/rpc_msg.h> +#include <glusterfs/compat.h> +#include <glusterfs/client_t.h> + +#ifndef MAX_IOVEC +#define MAX_IOVEC 16 +#endif + +/* TODO: we should store prognums at a centralized location to avoid conflict + or use a robust random number generator to avoid conflicts +*/ + +#define RPCSVC_INFRA_PROGRAM 7712846 /* random number */ + +typedef enum { + RPCSVC_PROC_EVENT_THREAD_DEATH = 0, +} rpcsvc_infra_procnum_t; + +#define RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT \ + 64 /* Default for protocol/server */ +#define RPCSVC_DEF_NFS_OUTSTANDING_RPC_LIMIT 16 /* Default for nfs/server */ +#define RPCSVC_MAX_OUTSTANDING_RPC_LIMIT 65536 +#define RPCSVC_MIN_OUTSTANDING_RPC_LIMIT 0 /* No limit i.e. Unlimited */ + +#define GF_RPCSVC "rpc-service" +#define RPCSVC_THREAD_STACK_SIZE ((size_t)(1024 * GF_UNIT_KB)) + +#define RPCSVC_FRAGHDR_SIZE 4 /* 4-byte RPC fragment header size */ +#define RPCSVC_DEFAULT_LISTEN_PORT GF_DEFAULT_BASE_PORT +#define RPCSVC_DEFAULT_MEMFACTOR 8 +#define RPCSVC_EVENTPOOL_SIZE_MULT 1024 +#define RPCSVC_POOLCOUNT_MULT 64 +#define RPCSVC_CONN_READ (128 * GF_UNIT_KB) +#define RPCSVC_PAGE_SIZE (128 * GF_UNIT_KB) +#define RPC_ROOT_UID 0 +#define RPC_ROOT_GID 0 +#define RPC_NOBODY_UID 65534 +#define RPC_NOBODY_GID 65534 + +/* RPC Record States */ +#define RPCSVC_READ_FRAGHDR 1 +#define RPCSVC_READ_FRAG 2 +/* The size in bytes, if crossed by a fragment will be handed over to the + * vectored actor so that it can allocate its buffers the way it wants. + * In our RPC layer, we assume that vectored RPC requests/records are never + * spread over multiple RPC fragments since that prevents us from determining + * whether the record should be handled in RPC layer completely or handed to + * the vectored handler. + */ +#define RPCSVC_VECTORED_FRAGSZ 4096 +#define RPCSVC_VECTOR_READCRED 1003 +#define RPCSVC_VECTOR_READVERFSZ 1004 +#define RPCSVC_VECTOR_READVERF 1005 +#define RPCSVC_VECTOR_IGNORE 1006 +#define RPCSVC_VECTOR_READVEC 1007 +#define RPCSVC_VECTOR_READPROCHDR 1008 + +#define rpcsvc_record_vectored_baremsg(rs) \ + (((rs)->state == RPCSVC_READ_FRAG) && (rs)->vecstate == 0) +#define rpcsvc_record_vectored_cred(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_READCRED) +#define rpcsvc_record_vectored_verfsz(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_READVERFSZ) +#define rpcsvc_record_vectored_verfread(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_READVERF) +#define rpcsvc_record_vectored_ignore(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_IGNORE) +#define rpcsvc_record_vectored_readvec(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_READVEC) +#define rpcsvc_record_vectored_readprochdr(rs) \ + ((rs)->vecstate == RPCSVC_VECTOR_READPROCHDR) +#define rpcsvc_record_vectored(rs) ((rs)->fragsize > RPCSVC_VECTORED_FRAGSZ) +/* Includes bytes up to and including the credential length field. The credlen + * will be followed by @credlen bytes of credential data which will have to be + * read separately by the vectored reader. After the credentials comes the + * verifier which will also have to be read separately including the 8 bytes of + * verf flavour and verflen. + */ +#define RPCSVC_BARERPC_MSGSZ 32 +#define rpcsvc_record_readfraghdr(rs) ((rs)->state == RPCSVC_READ_FRAGHDR) +#define rpcsvc_record_readfrag(rs) ((rs)->state == RPCSVC_READ_FRAG) + +#define RPCSVC_LOWVERS 2 +#define RPCSVC_HIGHVERS 2 + +#if 0 +#error "defined in /usr/include/rpc/auth.h" + +#define AUTH_NONE 0 /* no authentication */ +#define AUTH_NULL 0 /* backward compatibility */ +#define AUTH_SYS 1 /* unix style (uid, gids) */ +#define AUTH_UNIX AUTH_SYS +#define AUTH_SHORT 2 /* short hand unix style */ +#define AUTH_DES 3 /* des style (encrypted timestamps) */ +#define AUTH_DH AUTH_DES /* Diffie-Hellman (this is DES) */ +#define AUTH_KERB 4 /* kerberos style */ +#endif /* */ + +typedef struct rpcsvc_program rpcsvc_program_t; + +struct rpcsvc_notify_wrapper { + struct list_head list; + void *data; + rpcsvc_notify_t notify; +}; +typedef struct rpcsvc_notify_wrapper rpcsvc_notify_wrapper_t; + +typedef struct rpcsvc_request rpcsvc_request_t; + +typedef struct { + rpc_transport_t *trans; + rpcsvc_t *svc; + /* FIXME: remove address from this structure. Instead use get_myaddr + * interface implemented by individual transports. + */ + struct sockaddr_storage sa; + struct list_head list; +} rpcsvc_listener_t; + +struct rpcsvc_config { + int max_block_size; +}; + +#define rpcsvc_auth_flavour(au) ((au).flavour) + +typedef struct drc_client drc_client_t; +typedef struct drc_cached_op drc_cached_op_t; + +/* The container for the RPC call handed up to an actor. + * Dynamically allocated. Lives till the call reply is completely + * transmitted. + * */ +struct rpcsvc_request { + /* connection over which this request came. */ + rpc_transport_t *trans; + + rpcsvc_t *svc; + + rpcsvc_program_t *prog; + + int prognum; + + int progver; + + int procnum; + + int type; + + /* Uid and gid filled by the rpc-auth module during the authentication + * phase. + */ + uid_t uid; + gid_t gid; + pid_t pid; + + gf_lkowner_t lk_owner; + uint64_t gfs_id; + + /* Might want to move this to AUTH_UNIX specific state since this array + * is not available for every authentication scheme. + */ + gid_t *auxgids; + gid_t auxgidsmall[SMALL_GROUP_COUNT]; + gid_t *auxgidlarge; + int auxgidcount; + + /* The RPC message payload, contains the data required + * by the program actors. This is the buffer that will need to + * be de-xdred by the actor. + */ + int count; + struct iovec msg[MAX_IOVEC]; + + struct iobref *iobref; + + /* There can be cases of RPC requests where the reply needs to + * be built from multiple sources. E.g. where even the NFS reply + * can contain a payload, as in the NFSv3 read reply. Here the RPC header + * ,NFS header and the read data are brought together separately from + * different buffers, so we need to stage the buffers temporarily here + * before all of them get added to the connection's transmission list. + */ + struct list_head txlist; + + /* While the reply record is being built, this variable keeps track + * of how many bytes have been added to the record. + */ + size_t payloadsize; + + /* The credentials extracted from the rpc request */ + client_auth_data_t cred; + + /* The verified extracted from the rpc request. In request side + * processing this contains the verifier sent by the client, on reply + * side processing, it is filled with the verified that will be + * sent to the client. + */ + client_auth_data_t verf; + /* Container for a RPC program wanting to store a temp + * request-specific item. + */ + void *private; + + /* Container for transport to store request-specific item */ + void *trans_private; + + /* pointer to cached reply for use in DRC */ + drc_cached_op_t *reply; + + /* request queue in rpcsvc */ + struct list_head request_list; + + /* Status of the RPC call, whether it was accepted or denied. */ + int rpc_status; + + /* In case, the call was denied, the RPC error is stored here + * till the reply is sent. + */ + int rpc_err; + + /* In case the failure happened because of an authentication problem + * , this value needs to be assigned the correct auth error number. + */ + int auth_err; + + /* Things passed to rpc layer from client */ + + /* @flags: Can be used for binary data passed in xdata to be + passed here instead */ + unsigned int flags; + + /* ctime: origin of time on the client side, ideally this is + the one we should consider for time */ + struct timespec ctime; + + /* The identifier for the call from client. + * Needed to pair the reply with the call. + */ + uint32_t xid; + + /* Execute this request's actor function in ownthread of program?*/ + gf_boolean_t ownthread; + + gf_boolean_t synctask; + struct timespec begin; /*req handling start time*/ + struct timespec end; /*req handling end time*/ +}; + +#define rpcsvc_request_program(req) ((rpcsvc_program_t *)((req)->prog)) +#define rpcsvc_request_procnum(req) (((req)->procnum)) +#define rpcsvc_request_program_private(req) \ + (((rpcsvc_program_t *)((req)->prog))->private) +#define rpcsvc_request_accepted(req) ((req)->rpc_status == MSG_ACCEPTED) +#define rpcsvc_request_accepted_success(req) ((req)->rpc_err == SUCCESS) +#define rpcsvc_request_prog_minauth(req) (rpcsvc_request_program(req)->min_auth) +#define rpcsvc_request_cred_flavour(req) (rpcsvc_auth_flavour(req->cred)) +#define rpcsvc_request_verf_flavour(req) (rpcsvc_auth_flavour(req->verf)) +#define rpcsvc_request_service(req) ((req)->svc) +#define rpcsvc_request_uid(req) ((req)->uid) +#define rpcsvc_request_gid(req) ((req)->gid) +#define rpcsvc_request_private(req) ((req)->private) +#define rpcsvc_request_xid(req) ((req)->xid) +#define rpcsvc_request_set_private(req, prv) (req)->private = (void *)(prv) +#define rpcsvc_request_iobref_ref(req) (iobref_ref((req)->iobref)) +#define rpcsvc_request_record_ref(req) (iobuf_ref((req)->recordiob)) +#define rpcsvc_request_record_unref(req) (iobuf_unref((req)->recordiob)) +#define rpcsvc_request_record_iob(req) ((req)->recordiob) +#define rpcsvc_request_set_vecstate(req, state) ((req)->vecstate = state) +#define rpcsvc_request_vecstate(req) ((req)->vecstate) +#define rpcsvc_request_transport(req) ((req)->trans) +#define rpcsvc_request_transport_ref(req) (rpc_transport_ref((req)->trans)) +#define RPC_AUTH_ROOT_SQUASH(req) \ + do { \ + int gidcount = 0; \ + if (req->svc->root_squash) { \ + if (req->uid == RPC_ROOT_UID) \ + req->uid = req->svc->anonuid; \ + if (req->gid == RPC_ROOT_GID) \ + req->gid = req->svc->anongid; \ + \ + for (gidcount = 0; gidcount < req->auxgidcount; ++gidcount) { \ + if (!req->auxgids[gidcount]) \ + req->auxgids[gidcount] = req->svc->anongid; \ + } \ + } \ + } while (0); + +#define RPC_AUTH_ALL_SQUASH(req) \ + do { \ + int gidcount = 0; \ + if (req->svc->all_squash) { \ + req->uid = req->svc->anonuid; \ + req->gid = req->svc->anongid; \ + \ + for (gidcount = 0; gidcount < req->auxgidcount; ++gidcount) { \ + if (!req->auxgids[gidcount]) \ + req->auxgids[gidcount] = req->svc->anongid; \ + } \ + } \ + } while (0); + +#define RPCSVC_ACTOR_SUCCESS 0 +#define RPCSVC_ACTOR_ERROR (-1) +#define RPCSVC_ACTOR_IGNORE (-2) + +/* Functor for every type of protocol actor + * must be defined like this. + * + * See the request structure for info on how to handle the request + * in the program actor. + * + * On successful santify checks inside the actor, it should return + * RPCSVC_ACTOR_SUCCESS. + * On an error, on which the RPC layer is expected to return a reply, the actor + * should return RPCSVC_ACTOR_ERROR. + * + */ +typedef int (*rpcsvc_actor)(rpcsvc_request_t *req); +typedef int (*rpcsvc_vector_sizer)(int state, ssize_t *readsize, + char *base_addr, char *curr_addr); + +/* Every protocol actor will also need to specify the function the RPC layer + * will use to serialize or encode the message into XDR format just before + * transmitting on the connection. + */ +typedef void *(*rpcsvc_encode_reply)(void *msg); + +/* Once the reply has been transmitted, the message will have to be de-allocated + * , so every actor will need to provide a function that deallocates the message + * it had allocated as a response. + */ +typedef void (*rpcsvc_deallocate_reply)(void *msg); + +#define RPCSVC_NAME_MAX 32 +/* The descriptor for each procedure/actor that runs + * over the RPC service. + */ +typedef struct rpcsvc_actor_desc { + char procname[RPCSVC_NAME_MAX]; + rpcsvc_actor actor; + + /* Handler for cases where the RPC requests fragments are large enough + * to benefit from being decoded into aligned memory addresses. While + * decoding the request in a non-vectored manner, due to the nature of + * the XDR scheme, RPC cannot guarantee memory aligned addresses for + * the resulting message-specific structures. Allowing a specialized + * handler for letting the RPC program read the data from the network + * directly into its aligned buffers. + */ + rpcsvc_vector_sizer vector_sizer; + + int procnum; + + /* Can actor be ran on behalf an unprivileged requestor? */ + drc_op_type_t op_type; + gf_boolean_t unprivileged; +} rpcsvc_actor_t; + +typedef struct rpcsvc_request_queue { + struct list_head request_queue; + pthread_mutex_t queue_lock; + pthread_cond_t queue_cond; + pthread_t thread; + struct rpcsvc_program *program; + int gen; + gf_boolean_t waiting; +} rpcsvc_request_queue_t; + +/* Describes a program and its version along with the function pointers + * required to handle the procedures/actors of each program/version. + * Never changed ever by any thread so no need for a lock. + */ +struct rpcsvc_program { + char progname[RPCSVC_NAME_MAX]; + int prognum; + int progver; + /* FIXME */ + dict_t *options; /* An opaque dictionary + * populated by the program + * (probably from xl->options) + * which contain enough + * information for transport to + * initialize. As a part of + * cleanup, the members of + * options which are of interest + * to transport should be put + * into a structure for better + * readability and structure + * should replace options member + * here. + */ +#if 0 + int progaddrfamily; /* AF_INET or AF_INET6 */ + char *proghost; /* Bind host, can be NULL */ +#endif + rpcsvc_actor_t *actors; /* All procedure handlers */ + int numactors; /* Num actors in actor array */ + int proghighvers; /* Highest ver for program + supported by the system. */ + /* Program specific state handed to actors */ + void *private; + gf_latency_t *latencies; /*Tracks latency statistics for the rpc call*/ + + /* This upcall is provided by the program during registration. + * It is used to notify the program about events like connection being + * destroyed etc. The rpc program may take appropriate actions, for eg., + * in the case of connection being destroyed, it should cleanup its + * state stored in the connection. + */ + rpcsvc_notify_t notify; + + int proglowvers; /* Lowest ver */ + + /* An integer that identifies the min auth strength that is required + * by this protocol, for eg. MOUNT3 needs AUTH_UNIX at least. + * See RFC 1813, Section 5.2.1. + */ + int min_auth; + + /* list member to link to list of registered services with rpcsvc */ + struct list_head program; + rpcsvc_request_queue_t request_queue[EVENT_MAX_THREADS]; + pthread_mutex_t thr_lock; + pthread_cond_t thr_cond; + int threadcount; + int thr_queue; + pthread_key_t req_queue_key; + + /* eventthreadcount is just a readonly copy of the actual value + * owned by the event sub-system + * It is used to control the scaling of rpcsvc_request_handler threads + */ + int eventthreadcount; + uint16_t progport; /* Registered with portmap */ + /* Execute actor function in program's own thread? This will reduce */ + /* the workload on poller threads */ + gf_boolean_t ownthread; + gf_boolean_t alive; + + gf_boolean_t synctask; + unsigned long request_queue_status[EVENT_MAX_THREADS / __BITS_PER_LONG]; +}; + +typedef struct rpcsvc_cbk_program { + char *progname; + int prognum; + int progver; +} rpcsvc_cbk_program_t; +/* All users of RPC services should use this API to register their + * procedure handlers. + */ +extern int +rpcsvc_program_register(rpcsvc_t *svc, rpcsvc_program_t *program, + gf_boolean_t add_to_head); + +extern int +rpcsvc_program_unregister(rpcsvc_t *svc, rpcsvc_program_t *program); + +/* This will create and add a listener to listener pool. Programs can + * use any of the listener in this pool. A single listener can be used by + * multiple programs and vice versa. There can also be a one to one mapping + * between a program and a listener. After registering a program with rpcsvc, + * the program has to be associated with a listener using + * rpcsvc_program_register_portmap. + */ +/* FIXME: can multiple programs registered on same port? */ +extern int32_t +rpcsvc_create_listeners(rpcsvc_t *svc, dict_t *options, char *name); + +void +rpcsvc_listener_destroy(rpcsvc_listener_t *listener); + +extern int +rpcsvc_program_register_portmap(rpcsvc_program_t *newprog, uint32_t port); + +#ifdef IPV6_DEFAULT +extern int +rpcsvc_program_register_rpcbind6(rpcsvc_program_t *newprog, uint32_t port); +extern int +rpcsvc_program_unregister_rpcbind6(rpcsvc_program_t *newprog); +#endif + +extern int +rpcsvc_program_unregister_portmap(rpcsvc_program_t *newprog); + +extern int +rpcsvc_register_portmap_enabled(rpcsvc_t *svc); + +/* Inits the global RPC service data structures. + * Called in main. + */ +extern rpcsvc_t * +rpcsvc_init(xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options, + uint32_t poolcount); + +extern int +rpcsvc_reconfigure_options(rpcsvc_t *svc, dict_t *options); + +int +rpcsvc_register_notify(rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata); + +/* unregister a notification callback @notify with data @mydata from svc. + * returns the number of notification callbacks unregistered. + */ +int +rpcsvc_unregister_notify(rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata); + +int +rpcsvc_transport_submit(rpc_transport_t *trans, struct iovec *rpchdr, + int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *progpayload, + int progpayloadcount, struct iobref *iobref, + void *priv); + +int +rpcsvc_submit_message(rpcsvc_request_t *req, struct iovec *proghdr, + int hdrcount, struct iovec *payload, int payloadcount, + struct iobref *iobref); + +int +rpcsvc_submit_generic(rpcsvc_request_t *req, struct iovec *proghdr, + int hdrcount, struct iovec *payload, int payloadcount, + struct iobref *iobref); + +extern int +rpcsvc_error_reply(rpcsvc_request_t *req); + +#define RPCSVC_PEER_STRLEN 1024 +#define RPCSVC_AUTH_ACCEPT 1 +#define RPCSVC_AUTH_REJECT 2 +#define RPCSVC_AUTH_DONTCARE 3 + +extern int +rpcsvc_transport_peername(rpc_transport_t *trans, char *hostname, int hostlen); + +extern int +rpcsvc_transport_peeraddr(rpc_transport_t *trans, char *addrstr, int addrlen, + struct sockaddr_storage *returnsa, socklen_t sasize); + +extern int +rpcsvc_auth_check(rpcsvc_t *svc, char *volname, char *ipaddr); + +extern int +rpcsvc_transport_privport_check(rpcsvc_t *svc, char *volname, uint16_t port); + +#define rpcsvc_request_seterr(req, err) ((req)->rpc_err = (int)(err)) +#define rpcsvc_request_set_autherr(req, err) ((req)->auth_err = (int)(err)) + +extern int +rpcsvc_submit_vectors(rpcsvc_request_t *req); + +extern int +rpcsvc_request_attach_vector(rpcsvc_request_t *req, struct iovec msgvec, + struct iobuf *iob, struct iobref *ioref, + int finalvector); + +typedef int (*auth_init_trans)(rpc_transport_t *trans, void *priv); +typedef int (*auth_init_request)(rpcsvc_request_t *req, void *priv); +typedef int (*auth_request_authenticate)(rpcsvc_request_t *req, void *priv); + +/* This structure needs to be registered by every authentication scheme. + * Our authentication schemes are stored per connection because + * each connection will end up using a different authentication scheme. + */ +typedef struct rpcsvc_auth_ops { + auth_init_trans transport_init; + auth_init_request request_init; + auth_request_authenticate authenticate; +} rpcsvc_auth_ops_t; + +typedef struct rpcsvc_auth_flavour_desc { + char authname[RPCSVC_NAME_MAX]; + rpcsvc_auth_ops_t *authops; + void *authprivate; + int authnum; +} rpcsvc_auth_t; + +typedef void *(*rpcsvc_auth_initer_t)(rpcsvc_t *svc, dict_t *options); + +struct rpcsvc_auth_list { + struct list_head authlist; + rpcsvc_auth_initer_t init; + /* Should be the name with which we identify the auth scheme given + * in the volfile options. + * This should be different from the authname in rpc_auth_t + * in way that makes it easier to specify this scheme in the volfile. + * This is because the technical names of the schemes can be a bit + * arcane. + */ + char name[RPCSVC_NAME_MAX]; + rpcsvc_auth_t *auth; + int enable; +}; + +extern int +rpcsvc_auth_request_init(rpcsvc_request_t *req, struct rpc_msg *callmsg); + +extern int +rpcsvc_auth_init(rpcsvc_t *svc, dict_t *options); + +extern int +rpcsvc_auth_reconf(rpcsvc_t *svc, dict_t *options); + +extern int +rpcsvc_auth_transport_init(rpc_transport_t *xprt); + +extern int +rpcsvc_authenticate(rpcsvc_request_t *req); + +extern int +rpcsvc_auth_array(rpcsvc_t *svc, char *volname, int *autharr, int arrlen); + +/* If the request has been sent using AUTH_UNIX, this function returns the + * auxiliary gids as an array, otherwise, it returns NULL. + * Move to auth-unix specific source file when we need to modularize the + * authentication code even further to support mode auth schemes. + */ +extern gid_t * +rpcsvc_auth_unix_auxgids(rpcsvc_request_t *req, int *arrlen); + +extern char * +rpcsvc_volume_allowed(dict_t *options, char *volname); + +int +rpcsvc_request_submit(rpcsvc_t *rpc, rpc_transport_t *trans, + rpcsvc_cbk_program_t *prog, int procnum, void *req, + glusterfs_ctx_t *ctx, xdrproc_t xdrproc); + +int +rpcsvc_callback_submit(rpcsvc_t *rpc, rpc_transport_t *trans, + rpcsvc_cbk_program_t *prog, int procnum, + struct iovec *proghdr, int proghdrcount, + struct iobref *iobref); + +rpcsvc_actor_t * +rpcsvc_program_actor(rpcsvc_request_t *req); + +int +rpcsvc_transport_unix_options_build(dict_t *options, char *filepath); +int +rpcsvc_set_allow_insecure(rpcsvc_t *svc, dict_t *options); +int +rpcsvc_set_addr_namelookup(rpcsvc_t *svc, dict_t *options); +int +rpcsvc_set_root_squash(rpcsvc_t *svc, dict_t *options); +int +rpcsvc_set_all_squash(rpcsvc_t *svc, dict_t *options); +int +rpcsvc_set_outstanding_rpc_limit(rpcsvc_t *svc, dict_t *options, int defvalue); + +int +rpcsvc_set_throttle_on(rpcsvc_t *svc); + +int +rpcsvc_set_throttle_off(rpcsvc_t *svc); + +gf_boolean_t +rpcsvc_get_throttle(rpcsvc_t *svc); + +int +rpcsvc_auth_array(rpcsvc_t *svc, char *volname, int *autharr, int arrlen); +rpcsvc_vector_sizer +rpcsvc_get_program_vector_sizer(rpcsvc_t *svc, uint32_t prognum, + uint32_t progver, int procnum); +void +rpcsvc_autoscale_threads(glusterfs_ctx_t *ctx, rpcsvc_t *rpc, int incr); + +extern int +rpcsvc_destroy(rpcsvc_t *svc); +void +rpcsvc_statedump(rpcsvc_t *svc); +#endif diff --git a/rpc/rpc-lib/src/xdr-common.h b/rpc/rpc-lib/src/xdr-common.h new file mode 100644 index 00000000000..752736b3d4d --- /dev/null +++ b/rpc/rpc-lib/src/xdr-common.h @@ -0,0 +1,116 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _XDR_COMMON_H_ +#define _XDR_COMMON_H_ + +#include <rpc/types.h> +#include <sys/types.h> +#include <rpc/xdr.h> +#include <rpc/auth.h> +#include <sys/uio.h> + +#ifdef __NetBSD__ +#include <dirent.h> +#endif /* __NetBSD__ */ + +enum gf_dump_procnum { + GF_DUMP_NULL, + GF_DUMP_DUMP, + GF_DUMP_PING, + GF_DUMP_MAXVALUE, +}; + +#define GLUSTER_DUMP_PROGRAM 123451501 /* Completely random */ +#define GLUSTER_DUMP_VERSION 1 + +/* MAX_AUTH_BYTES is restricted to 400 bytes, see + * http://tools.ietf.org/html/rfc5531#section-8.2 */ +#define GF_MAX_AUTH_BYTES MAX_AUTH_BYTES + +/* The size of an AUTH_GLUSTERFS_V2 structure: + * + * 1 | pid + * 1 | uid + * 1 | gid + * 1 | groups_len + * XX | groups_val (GF_MAX_AUX_GROUPS=65535) + * 1 | lk_owner_len + * YY | lk_owner_val (GF_MAX_LOCK_OWNER_LEN=1024) + * ----+------------------------------------------- + * 5 | total xdr-units + * + * one XDR-unit is defined as BYTES_PER_XDR_UNIT = 4 bytes + * MAX_AUTH_BYTES = 400 is the maximum, this is 100 xdr-units. + * XX + YY can be 95 to fill the 100 xdr-units. + * + * Note that the on-wire protocol has tighter requirements than the internal + * structures. It is possible for xlators to use more groups and a bigger + * lk_owner than that can be sent by a GlusterFS-client. + * + * ------- + * On v3, there are 4 more units, and hence it will be 9 xdr-units + */ +#define GF_AUTH_GLUSTERFS_MAX_GROUPS(lk_len, type) \ + ((type == AUTH_GLUSTERFS_v2) ? (95 - lk_len) : (91 - lk_len)) +#define GF_AUTH_GLUSTERFS_MAX_LKOWNER(groups_len, type) \ + ((type == AUTH_GLUSTERFS_v2) ? (95 - groups_len) : (91 - groups_len)) + +#ifdef GF_LINUX_HOST_OS +#define xdr_u_int32_t xdr_uint32_t +#define xdr_u_int64_t xdr_uint64_t +unsigned long +xdr_sizeof(xdrproc_t func, void *data); +#endif + +#ifdef GF_DARWIN_HOST_OS +#define xdr_u_quad_t xdr_u_int64_t +#define xdr_quad_t xdr_int64_t +#define xdr_uint32_t xdr_u_int32_t +#define xdr_uint64_t xdr_u_int64_t +#define uint64_t u_int64_t +unsigned long +xdr_sizeof(xdrproc_t func, void *data); +#endif + +#if defined(__NetBSD__) +#define xdr_u_quad_t xdr_u_int64_t +#define xdr_quad_t xdr_int64_t +#define xdr_uint32_t xdr_u_int32_t +#define xdr_uint64_t xdr_u_int64_t +#endif + +#ifdef GF_SOLARIS_HOST_OS +#define u_quad_t uint64_t +#define quad_t int64_t +#define xdr_u_quad_t xdr_uint64_t +#define xdr_quad_t xdr_int64_t +#endif + +/* Returns the address of the byte that follows the + * last byte used for decoding the previous xdr component. + * E.g. once the RPC call for NFS has been decoded, the macro will return + * the address from which the NFS header starts. + */ +#define xdr_decoded_remaining_addr(xdr) ((&xdr)->x_private) + +/* Returns the length of the remaining record after the previous decode + * operation completed. + */ +#define xdr_decoded_remaining_len(xdr) ((&xdr)->x_handy) + +/* Returns the number of bytes used by the last encode operation. */ +#define xdr_encoded_length(xdr) \ + (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)) + +#define xdr_decoded_length(xdr) \ + (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)) + +#endif diff --git a/rpc/rpc-lib/src/xdr-rpc.c b/rpc/rpc-lib/src/xdr-rpc.c new file mode 100644 index 00000000000..4992dc5a7ce --- /dev/null +++ b/rpc/rpc-lib/src/xdr-rpc.c @@ -0,0 +1,198 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include <rpc/rpc.h> +#include <rpc/xdr.h> +#include <sys/uio.h> +#include <rpc/auth_unix.h> + +#include "xdr-rpc.h" +#include "xdr-common.h" +#include <glusterfs/common-utils.h> + +/* Decodes the XDR format in msgbuf into rpc_msg. + * The remaining payload is returned into payload. + */ +int +xdr_to_rpc_call(char *msgbuf, size_t len, struct rpc_msg *call, + struct iovec *payload, char *credbytes, char *verfbytes) +{ + XDR xdr; + char opaquebytes[GF_MAX_AUTH_BYTES]; + struct opaque_auth *oa = NULL; + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", msgbuf, out); + GF_VALIDATE_OR_GOTO("rpc", call, out); + + memset(call, 0, sizeof(*call)); + + oa = &call->rm_call.cb_cred; + if (!credbytes) + oa->oa_base = opaquebytes; + else + oa->oa_base = credbytes; + + oa = &call->rm_call.cb_verf; + if (!verfbytes) + oa->oa_base = opaquebytes; + else + oa->oa_base = verfbytes; + + xdrmem_create(&xdr, msgbuf, len, XDR_DECODE); + if (!xdr_callmsg(&xdr, call)) { + gf_log("rpc", GF_LOG_WARNING, "failed to decode call msg"); + goto out; + } + + if (payload) { + payload->iov_base = xdr_decoded_remaining_addr(xdr); + payload->iov_len = xdr_decoded_remaining_len(xdr); + } + + ret = 0; +out: + return ret; +} + +bool_t +true_func(XDR *s, caddr_t *a) +{ + return TRUE; +} + +int +rpc_fill_empty_reply(struct rpc_msg *reply, uint32_t xid) +{ + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", reply, out); + + /* Setting to 0 also results in reply verifier flavor to be + * set to AUTH_NULL which is what we want right now. + */ + memset(reply, 0, sizeof(*reply)); + reply->rm_xid = xid; + reply->rm_direction = REPLY; + + ret = 0; +out: + return ret; +} + +int +rpc_fill_denied_reply(struct rpc_msg *reply, int rjstat, int auth_err) +{ + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", reply, out); + + reply->rm_reply.rp_stat = MSG_DENIED; + reply->rjcted_rply.rj_stat = rjstat; + if (rjstat == RPC_MISMATCH) { + /* No problem with hardcoding + * RPC version numbers. We only support + * v2 anyway. + */ + reply->rjcted_rply.rj_vers.low = 2; + reply->rjcted_rply.rj_vers.high = 2; + } else if (rjstat == AUTH_ERROR) + reply->rjcted_rply.rj_why = auth_err; + + ret = 0; +out: + return ret; +} + +int +rpc_fill_accepted_reply(struct rpc_msg *reply, int arstat, int proglow, + int proghigh, int verf, int len, char *vdata) +{ + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", reply, out); + + reply->rm_reply.rp_stat = MSG_ACCEPTED; + reply->acpted_rply.ar_stat = arstat; + + reply->acpted_rply.ar_verf.oa_flavor = verf; + reply->acpted_rply.ar_verf.oa_length = len; + reply->acpted_rply.ar_verf.oa_base = vdata; + if (arstat == PROG_MISMATCH) { + reply->acpted_rply.ar_vers.low = proglow; + reply->acpted_rply.ar_vers.high = proghigh; + } else if (arstat == SUCCESS) { + /* This is a hack. I'd really like to build a custom + * XDR library because Sun RPC interface is not very flexible. + */ + reply->acpted_rply.ar_results.proc = (xdrproc_t)true_func; + reply->acpted_rply.ar_results.where = NULL; + } + + ret = 0; +out: + return ret; +} + +int +rpc_reply_to_xdr(struct rpc_msg *reply, char *dest, size_t len, + struct iovec *dst) +{ + XDR xdr; + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", reply, out); + GF_VALIDATE_OR_GOTO("rpc", dest, out); + GF_VALIDATE_OR_GOTO("rpc", dst, out); + + xdrmem_create(&xdr, dest, len, XDR_ENCODE); + if (!xdr_replymsg(&xdr, reply)) { + gf_log("rpc", GF_LOG_WARNING, "failed to encode reply msg"); + goto out; + } + + dst->iov_base = dest; + dst->iov_len = xdr_encoded_length(xdr); + + ret = 0; +out: + return ret; +} + +int +xdr_to_auth_unix_cred(char *msgbuf, int msglen, struct authunix_parms *au, + char *machname, gid_t *gids) +{ + XDR xdr; + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", msgbuf, out); + GF_VALIDATE_OR_GOTO("rpc", machname, out); + GF_VALIDATE_OR_GOTO("rpc", gids, out); + GF_VALIDATE_OR_GOTO("rpc", au, out); + + au->aup_machname = machname; +#if defined(GF_DARWIN_HOST_OS) || defined(__FreeBSD__) + au->aup_gids = (int *)gids; +#else + au->aup_gids = gids; +#endif + + xdrmem_create(&xdr, msgbuf, msglen, XDR_DECODE); + + if (!xdr_authunix_parms(&xdr, au)) { + gf_log("rpc", GF_LOG_WARNING, "failed to decode auth unix parms"); + goto out; + } + + ret = 0; +out: + return ret; +} diff --git a/rpc/rpc-lib/src/xdr-rpc.h b/rpc/rpc-lib/src/xdr-rpc.h new file mode 100644 index 00000000000..7baed273846 --- /dev/null +++ b/rpc/rpc-lib/src/xdr-rpc.h @@ -0,0 +1,95 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _XDR_RPC_H_ +#define _XDR_RPC_H_ + +#ifndef GF_SOLARIS_HOST_OS +#include <rpc/rpc.h> +#endif + +#ifdef GF_SOLARIS_HOST_OS +#include <rpc/auth.h> +#include <rpc/auth_sys.h> +#endif + +#include <arpa/inet.h> +#include <rpc/xdr.h> +#include <sys/uio.h> + +#include "xdr-common.h" + +typedef enum { + AUTH_GLUSTERFS = 5, + AUTH_GLUSTERFS_v2 = 390039, /* using a number from 'unused' range, + from the list available in RFC5531 */ + AUTH_GLUSTERFS_v3 = 390040, /* this too is unused */ +} gf_rpc_authtype_t; + +/* Converts a given network buffer from its XDR format to a structure + * that contains everything an RPC call needs to work. + */ +extern int +xdr_to_rpc_call(char *msgbuf, size_t len, struct rpc_msg *call, + struct iovec *payload, char *credbytes, char *verfbytes); + +extern int +rpc_fill_empty_reply(struct rpc_msg *reply, uint32_t xid); + +extern int +rpc_fill_denied_reply(struct rpc_msg *reply, int rjstat, int auth_err); + +extern int +rpc_fill_accepted_reply(struct rpc_msg *reply, int arstat, int proglow, + int proghigh, int verf, int len, char *vdata); +extern int +rpc_reply_to_xdr(struct rpc_msg *reply, char *dest, size_t len, + struct iovec *dst); + +extern int +xdr_to_auth_unix_cred(char *msgbuf, int msglen, struct authunix_parms *au, + char *machname, gid_t *gids); +/* Macros that simplify accessing the members of an RPC call structure. */ +#define rpc_call_xid(call) ((call)->rm_xid) +#define rpc_call_direction(call) ((call)->rm_direction) +#define rpc_call_rpcvers(call) ((call)->ru.RM_cmb.cb_rpcvers) +#define rpc_call_program(call) ((call)->ru.RM_cmb.cb_prog) +#define rpc_call_progver(call) ((call)->ru.RM_cmb.cb_vers) +#define rpc_call_progproc(call) ((call)->ru.RM_cmb.cb_proc) +#define rpc_opaque_auth_flavour(oa) ((oa)->oa_flavor) +#define rpc_opaque_auth_len(oa) ((oa)->oa_length) + +#define rpc_call_cred_flavour(call) \ + (rpc_opaque_auth_flavour((&(call)->ru.RM_cmb.cb_cred))) +#define rpc_call_cred_len(call) \ + (rpc_opaque_auth_len((&(call)->ru.RM_cmb.cb_cred))) + +#define rpc_call_verf_flavour(call) \ + (rpc_opaque_auth_flavour((&(call)->ru.RM_cmb.cb_verf))) +#define rpc_call_verf_len(call) \ + (rpc_opaque_auth_len((&(call)->ru.RM_cmb.cb_verf))) + +#if defined(GF_DARWIN_HOST_OS) || !defined(HAVE_RPC_RPC_H) +#define GF_PRI_RPC_XID PRIu32 +#define GF_PRI_RPC_VERSION PRIu32 +#define GF_PRI_RPC_PROG_ID PRIu32 +#define GF_PRI_RPC_PROG_VERS PRIu32 +#define GF_PRI_RPC_PROC PRIu32 +#define GF_PRI_RPC_PROC_VERSION PRIu32 +#else +#define GF_PRI_RPC_XID PRIu64 +#define GF_PRI_RPC_VERSION PRIu64 +#define GF_PRI_RPC_PROG_ID PRIu64 +#define GF_PRI_RPC_PROG_VERS PRIu64 +#define GF_PRI_RPC_PROC PRIu64 +#define GF_PRI_RPC_PROC_VERSION PRIu64 +#endif + +#endif diff --git a/rpc/rpc-lib/src/xdr-rpcclnt.c b/rpc/rpc-lib/src/xdr-rpcclnt.c new file mode 100644 index 00000000000..8dcdcfeda83 --- /dev/null +++ b/rpc/rpc-lib/src/xdr-rpcclnt.c @@ -0,0 +1,105 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include <rpc/rpc.h> +#include <rpc/xdr.h> +#include <sys/uio.h> +#include <rpc/auth_unix.h> +#include <errno.h> + +#include "xdr-rpc.h" +#include "xdr-common.h" +#include <glusterfs/common-utils.h> + +/* Decodes the XDR format in msgbuf into rpc_msg. + * The remaining payload is returned into payload. + */ +int +xdr_to_rpc_reply(char *msgbuf, size_t len, struct rpc_msg *reply, + struct iovec *payload, char *verfbytes) +{ + XDR xdr; + int ret = -EINVAL; + + GF_VALIDATE_OR_GOTO("rpc", msgbuf, out); + GF_VALIDATE_OR_GOTO("rpc", reply, out); + + memset(reply, 0, sizeof(struct rpc_msg)); + + reply->acpted_rply.ar_verf = _null_auth; + reply->acpted_rply.ar_results.where = NULL; + reply->acpted_rply.ar_results.proc = (xdrproc_t)(xdr_void); + + xdrmem_create(&xdr, msgbuf, len, XDR_DECODE); + if (!xdr_replymsg(&xdr, reply)) { + gf_log("rpc", GF_LOG_WARNING, "failed to decode reply msg"); + goto out; + } + if (payload) { + payload->iov_base = xdr_decoded_remaining_addr(xdr); + payload->iov_len = xdr_decoded_remaining_len(xdr); + } + + ret = 0; +out: + return ret; +} + +int +rpc_request_to_xdr(struct rpc_msg *request, char *dest, size_t len, + struct iovec *dst) +{ + XDR xdr; + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", dest, out); + GF_VALIDATE_OR_GOTO("rpc", request, out); + GF_VALIDATE_OR_GOTO("rpc", dst, out); + + xdrmem_create(&xdr, dest, len, XDR_ENCODE); + if (!xdr_callmsg(&xdr, request)) { + gf_log("rpc", GF_LOG_WARNING, "failed to encode call msg"); + goto out; + } + + dst->iov_base = dest; + dst->iov_len = xdr_encoded_length(xdr); + + ret = 0; + +out: + return ret; +} + +int +auth_unix_cred_to_xdr(struct authunix_parms *au, char *dest, size_t len, + struct iovec *iov) +{ + XDR xdr; + int ret = -1; + + GF_VALIDATE_OR_GOTO("rpc", au, out); + GF_VALIDATE_OR_GOTO("rpc", dest, out); + GF_VALIDATE_OR_GOTO("rpc", iov, out); + + xdrmem_create(&xdr, dest, len, XDR_DECODE); + + if (!xdr_authunix_parms(&xdr, au)) { + gf_log("rpc", GF_LOG_WARNING, "failed to decode authunix parms"); + goto out; + } + + iov->iov_base = dest; + iov->iov_len = xdr_encoded_length(xdr); + + ret = 0; +out: + return ret; +} diff --git a/rpc/rpc-lib/src/xdr-rpcclnt.h b/rpc/rpc-lib/src/xdr-rpcclnt.h new file mode 100644 index 00000000000..58eda4892a9 --- /dev/null +++ b/rpc/rpc-lib/src/xdr-rpcclnt.h @@ -0,0 +1,36 @@ +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#ifndef _XDR_RPCCLNT_H +#define _XDR_RPCCLNT_H + +#include <arpa/inet.h> +#include <rpc/xdr.h> +#include <sys/uio.h> +#include <rpc/rpc_msg.h> +#include <rpc/auth_unix.h> + +/* Macros that simplify accessing the members of an RPC call structure. */ +#define rpc_reply_xid(reply) ((reply)->rm_xid) +#define rpc_reply_status(reply) ((reply)->ru.RM_rmb.rp_stat) +#define rpc_accepted_reply_status(reply) ((reply)->acpted_rply.ar_stat) +#define rpc_reply_verf_flavour(reply) ((reply)->acpted_rply.ar_verf.oa_flavor) + +int +xdr_to_rpc_reply(char *msgbuf, size_t len, struct rpc_msg *reply, + struct iovec *payload, char *verfbytes); +int +rpc_request_to_xdr(struct rpc_msg *request, char *dest, size_t len, + struct iovec *dst); +int +auth_unix_cred_to_xdr(struct authunix_parms *au, char *dest, size_t len, + struct iovec *iov); + +#endif |