From: Ben Pfaff Date: Wed, 17 Jun 2009 21:35:35 +0000 (-0700) Subject: dpif: Make dpifs abstract, to allow multiple datapath implementations. X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=96fba48f52254c0cef942dcce130e33d290297da;p=openvswitch dpif: Make dpifs abstract, to allow multiple datapath implementations. This commit initially introduces only a single datapath implementation, which is the same as the original one, but it paves the way for additional implementations, such as the upcoming userspace datapath. --- diff --git a/include/openvswitch/datapath-protocol.h b/include/openvswitch/datapath-protocol.h index 9b601a3c..44875881 100644 --- a/include/openvswitch/datapath-protocol.h +++ b/include/openvswitch/datapath-protocol.h @@ -70,13 +70,12 @@ #define ODP_PORT_GROUP_GET _IOWR('O', 12, struct odp_port_group) #define ODP_FLOW_GET _IOWR('O', 13, struct odp_flow) +#define ODP_FLOW_PUT _IOWR('O', 14, struct odp_flow) #define ODP_FLOW_LIST _IOWR('O', 15, struct odp_flowvec) - #define ODP_FLOW_FLUSH _IO('O', 16) -#define ODP_FLOW_PUT _IOWR('O', 17, struct odp_flow) -#define ODP_FLOW_DEL _IOWR('O', 18, struct odp_flow) +#define ODP_FLOW_DEL _IOWR('O', 17, struct odp_flow) -#define ODP_EXECUTE _IOR('O', 19, struct odp_execute) +#define ODP_EXECUTE _IOR('O', 18, struct odp_execute) struct odp_stats { /* Flows. */ diff --git a/lib/automake.mk b/lib/automake.mk index f6798a1e..54b3c60e 100644 --- a/lib/automake.mk +++ b/lib/automake.mk @@ -32,6 +32,10 @@ lib_libopenvswitch_a_SOURCES = \ lib/dhcp.h \ lib/dhparams.h \ lib/dirs.h \ + lib/dpif-linux.c \ + lib/dpif-provider.h \ + lib/dpif.c \ + lib/dpif.h \ lib/dynamic-string.c \ lib/dynamic-string.h \ lib/fatal-signal.c \ @@ -117,8 +121,6 @@ CLEANFILES += $(nodist_lib_libopenvswitch_a_SOURCES) if HAVE_NETLINK lib_libopenvswitch_a_SOURCES += \ - lib/dpif.c \ - lib/dpif.h \ lib/netlink-protocol.h \ lib/netlink.c \ lib/netlink.h diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c new file mode 100644 index 00000000..2c8c7b09 --- /dev/null +++ b/lib/dpif-linux.c @@ -0,0 +1,603 @@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "dpif.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "dpif-provider.h" +#include "ofpbuf.h" +#include "poll-loop.h" +#include "util.h" + +#include "vlog.h" +#define THIS_MODULE VLM_dpif_linux + +/* Datapath interface for the openvswitch Linux kernel module. */ +struct dpif_linux { + struct dpif dpif; + int fd; +}; + +static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); + +static int do_ioctl(const struct dpif *, int cmd, const void *arg); +static int lookup_minor(const char *name, int *minor); +static int create_minor(const char *name, int minor, struct dpif **dpifp); +static int open_minor(int minor, struct dpif **dpifp); +static int make_openvswitch_device(int minor, char **fnp); + +static struct dpif_linux * +dpif_linux_cast(const struct dpif *dpif) +{ + dpif_assert_class(dpif, &dpif_linux_class); + return CONTAINER_OF(dpif, struct dpif_linux, dpif); +} + +static int +dpif_linux_open(const char *name UNUSED, char *suffix, bool create, + struct dpif **dpifp) +{ + int minor; + + minor = !strncmp(name, "dp", 2) && isdigit(name[2]) ? atoi(name + 2) : -1; + if (create) { + if (minor >= 0) { + return create_minor(suffix, minor, dpifp); + } else { + /* Scan for unused minor number. */ + for (minor = 0; minor < ODP_MAX; minor++) { + int error = create_minor(suffix, minor, dpifp); + if (error != EBUSY) { + return error; + } + } + + /* All datapath numbers in use. */ + return ENOBUFS; + } + } else { + struct dpif_linux *dpif; + int listen_mask; + int error; + + if (minor < 0) { + error = lookup_minor(suffix, &minor); + if (error) { + return error; + } + } + + error = open_minor(minor, dpifp); + if (error) { + return error; + } + dpif = dpif_linux_cast(*dpifp); + + /* We can open the device, but that doesn't mean that it's been + * created. If it hasn't been, then any command other than + * ODP_DP_CREATE will return ENODEV. Try something innocuous. */ + listen_mask = 0; /* Make Valgrind happy. */ + error = do_ioctl(*dpifp, ODP_GET_LISTEN_MASK, &listen_mask); + if (error) { + if (error != ENODEV) { + VLOG_WARN("%s: probe returned unexpected error: %s", + dpif_name(*dpifp), strerror(error)); + } + dpif_close(*dpifp); + } + return error; + } +} + +static void +dpif_linux_close(struct dpif *dpif_) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + close(dpif->fd); + free(dpif); +} + +static int +dpif_linux_delete(struct dpif *dpif_) +{ + return do_ioctl(dpif_, ODP_DP_DESTROY, NULL); +} + +static int +dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) +{ + return do_ioctl(dpif_, ODP_DP_STATS, stats); +} + +static int +dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) +{ + int drop_frags; + int error; + + error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags); + if (!error) { + *drop_fragsp = drop_frags & 1; + } + return error; +} + +static int +dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) +{ + int drop_frags_int = drop_frags; + return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int); +} + +static int +dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags, + uint16_t *port_no) +{ + struct odp_port port; + int error; + + memset(&port, 0, sizeof port); + strncpy(port.devname, devname, sizeof port.devname); + port.flags = flags; + error = do_ioctl(dpif_, ODP_PORT_ADD, &port); + if (!error) { + *port_no = port.port; + } + return error; +} + +static int +dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) +{ + int tmp = port_no; + return do_ioctl(dpif_, ODP_PORT_DEL, &tmp); +} + +static int +dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no, + struct odp_port *port) +{ + memset(port, 0, sizeof *port); + port->port = port_no; + return do_ioctl(dpif_, ODP_PORT_QUERY, port); +} + +static int +dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname, + struct odp_port *port) +{ + memset(port, 0, sizeof *port); + strncpy(port->devname, devname, sizeof port->devname); + return do_ioctl(dpif_, ODP_PORT_QUERY, port); +} + +static int +dpif_linux_flow_flush(struct dpif *dpif_) +{ + return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL); +} + +static int +dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n) +{ + struct odp_portvec pv; + int error; + + pv.ports = ports; + pv.n_ports = n; + error = do_ioctl(dpif_, ODP_PORT_LIST, &pv); + return error ? -error : pv.n_ports; +} + +static int +dpif_linux_port_group_get(const struct dpif *dpif_, int group, + uint16_t ports[], int n) +{ + struct odp_port_group pg; + int error; + + assert(n <= UINT16_MAX); + pg.group = group; + pg.ports = ports; + pg.n_ports = n; + error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg); + return error ? -error : pg.n_ports; +} + +static int +dpif_linux_port_group_set(struct dpif *dpif_, int group, + const uint16_t ports[], int n) +{ + struct odp_port_group pg; + + assert(n <= UINT16_MAX); + pg.group = group; + pg.ports = (uint16_t *) ports; + pg.n_ports = n; + return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg); +} + +static int +dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n) +{ + struct odp_flowvec fv; + fv.flows = flows; + fv.n_flows = n; + return do_ioctl(dpif_, ODP_FLOW_GET, &fv); +} + +static int +dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put) +{ + return do_ioctl(dpif_, ODP_FLOW_PUT, put); +} + +static int +dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow) +{ + return do_ioctl(dpif_, ODP_FLOW_DEL, flow); +} + +static int +dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n) +{ + struct odp_flowvec fv; + int error; + + fv.flows = flows; + fv.n_flows = n; + error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv); + return error ? -error : fv.n_flows; +} + +static int +dpif_linux_execute(struct dpif *dpif_, uint16_t in_port, + const union odp_action actions[], int n_actions, + const struct ofpbuf *buf) +{ + struct odp_execute execute; + memset(&execute, 0, sizeof execute); + execute.in_port = in_port; + execute.actions = (union odp_action *) actions; + execute.n_actions = n_actions; + execute.data = buf->data; + execute.length = buf->size; + return do_ioctl(dpif_, ODP_EXECUTE, &execute); +} + +static int +dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask) +{ + return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask); +} + +static int +dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask) +{ + return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask); +} + +static int +dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct ofpbuf *buf; + int retval; + int error; + + buf = ofpbuf_new(65536); + retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); + if (retval < 0) { + error = errno; + if (error != EAGAIN) { + VLOG_WARN_RL(&error_rl, "%s: read failed: %s", + dpif_name(dpif_), strerror(error)); + } + } else if (retval >= sizeof(struct odp_msg)) { + struct odp_msg *msg = buf->data; + if (msg->length <= retval) { + buf->size += retval; + *bufp = buf; + return 0; + } else { + VLOG_WARN_RL(&error_rl, "%s: discarding message truncated " + "from %zu bytes to %d", + dpif_name(dpif_), msg->length, retval); + error = ERANGE; + } + } else if (!retval) { + VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_)); + error = EPROTO; + } else { + VLOG_WARN_RL(&error_rl, + "%s: discarding too-short message (%d bytes)", + dpif_name(dpif_), retval); + error = ERANGE; + } + + *bufp = NULL; + ofpbuf_delete(buf); + return error; +} + +static void +dpif_linux_recv_wait(struct dpif *dpif_) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + poll_fd_wait(dpif->fd, POLLIN); +} + +const struct dpif_class dpif_linux_class = { + "", /* This is the default class. */ + "linux", + dpif_linux_open, + dpif_linux_close, + dpif_linux_delete, + dpif_linux_get_stats, + dpif_linux_get_drop_frags, + dpif_linux_set_drop_frags, + dpif_linux_port_add, + dpif_linux_port_del, + dpif_linux_port_query_by_number, + dpif_linux_port_query_by_name, + dpif_linux_port_list, + dpif_linux_port_group_get, + dpif_linux_port_group_set, + dpif_linux_flow_get, + dpif_linux_flow_put, + dpif_linux_flow_del, + dpif_linux_flow_flush, + dpif_linux_flow_list, + dpif_linux_execute, + dpif_linux_recv_get_mask, + dpif_linux_recv_set_mask, + dpif_linux_recv, + dpif_linux_recv_wait, +}; + +static int get_openvswitch_major(void); +static int get_major(const char *target, int default_major); + +static int +do_ioctl(const struct dpif *dpif_, int cmd, const void *arg) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + return ioctl(dpif->fd, cmd, arg) ? errno : 0; +} + +static int +lookup_minor(const char *name, int *minor) +{ + struct ethtool_drvinfo drvinfo; + struct ifreq ifr; + int error; + int sock; + + sock = socket(AF_INET, SOCK_DGRAM, 0); + if (sock < 0) { + VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno)); + error = errno; + goto error; + } + + memset(&ifr, 0, sizeof ifr); + strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + ifr.ifr_data = (caddr_t) &drvinfo; + + memset(&drvinfo, 0, sizeof drvinfo); + drvinfo.cmd = ETHTOOL_GDRVINFO; + if (ioctl(sock, SIOCETHTOOL, &ifr)) { + VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno)); + error = errno; + goto error_close_sock; + } + + if (strcmp(drvinfo.driver, "openvswitch")) { + VLOG_WARN("%s is not an openvswitch device", name); + error = EOPNOTSUPP; + goto error_close_sock; + } + + if (!isdigit(drvinfo.bus_info[0])) { + VLOG_WARN("%s ethtool info does not contain an openvswitch minor", + name); + error = EPROTOTYPE; + goto error_close_sock; + } + + *minor = atoi(drvinfo.bus_info); + close(sock); + return 0; + +error_close_sock: + close(sock); +error: + return error; +} + +static int +make_openvswitch_device(int minor, char **fnp) +{ + dev_t dev = makedev(get_openvswitch_major(), minor); + const char dirname[] = "/dev/net"; + struct stat s; + char fn[128]; + + *fnp = NULL; + sprintf(fn, "%s/dp%d", dirname, minor); + if (!stat(fn, &s)) { + if (!S_ISCHR(s.st_mode)) { + VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing", + fn); + } else if (s.st_rdev != dev) { + VLOG_WARN_RL(&error_rl, + "%s is device %u:%u instead of %u:%u, fixing", + fn, major(s.st_rdev), minor(s.st_rdev), + major(dev), minor(dev)); + } else { + goto success; + } + if (unlink(fn)) { + VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)", + fn, strerror(errno)); + return errno; + } + } else if (errno == ENOENT) { + if (stat(dirname, &s)) { + if (errno == ENOENT) { + if (mkdir(dirname, 0755)) { + VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)", + dirname, strerror(errno)); + return errno; + } + } else { + VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", + dirname, strerror(errno)); + return errno; + } + } + } else { + VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno)); + return errno; + } + + /* The device needs to be created. */ + if (mknod(fn, S_IFCHR | 0700, dev)) { + VLOG_WARN_RL(&error_rl, + "%s: creating character device %u:%u failed (%s)", + fn, major(dev), minor(dev), strerror(errno)); + return errno; + } + +success: + *fnp = xstrdup(fn); + return 0; +} + + +static int +get_openvswitch_major(void) +{ + static unsigned int openvswitch_major; + if (!openvswitch_major) { + enum { DEFAULT_MAJOR = 248 }; + openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR); + } + return openvswitch_major; +} + +static int +get_major(const char *target, int default_major) +{ + const char fn[] = "/proc/devices"; + char line[128]; + FILE *file; + int ln; + + file = fopen(fn, "r"); + if (!file) { + VLOG_ERR("opening %s failed (%s)", fn, strerror(errno)); + goto error; + } + + for (ln = 1; fgets(line, sizeof line, file); ln++) { + char name[64]; + int major; + + if (!strncmp(line, "Character", 9) || line[0] == '\0') { + /* Nothing to do. */ + } else if (!strncmp(line, "Block", 5)) { + /* We only want character devices, so skip the rest of the file. */ + break; + } else if (sscanf(line, "%d %63s", &major, name)) { + if (!strcmp(name, target)) { + fclose(file); + return major; + } + } else { + static bool warned; + if (!warned) { + VLOG_WARN("%s:%d: syntax error", fn, ln); + } + warned = true; + } + } + + VLOG_ERR("%s: %s major not found (is the module loaded?), using " + "default major %d", fn, target, default_major); +error: + VLOG_INFO("using default major %d for %s", default_major, target); + return default_major; +} + +static int +create_minor(const char *name, int minor, struct dpif **dpifp) +{ + int error = open_minor(minor, dpifp); + if (!error) { + error = do_ioctl(*dpifp, ODP_DP_CREATE, name); + if (error) { + dpif_close(*dpifp); + } + } + return error; +} + +static int +open_minor(int minor, struct dpif **dpifp) +{ + int error; + char *fn; + int fd; + + error = make_openvswitch_device(minor, &fn); + if (error) { + return error; + } + + fd = open(fn, O_RDONLY | O_NONBLOCK); + if (fd >= 0) { + struct dpif_linux *dpif; + char *name; + + name = xasprintf("dp%d", minor); + + dpif = xmalloc(sizeof *dpif); + dpif_init(&dpif->dpif, &dpif_linux_class, name, minor, minor); + dpif->fd = fd; + *dpifp = &dpif->dpif; + + free(name); + } else { + error = errno; + VLOG_WARN("%s: open failed (%s)", fn, strerror(error)); + } + free(fn); + + return error; +} diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h new file mode 100644 index 00000000..902064c5 --- /dev/null +++ b/lib/dpif-provider.h @@ -0,0 +1,237 @@ +/* + * Copyright (c) 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef DPIF_PROVIDER_H +#define DPIF_PROVIDER_H 1 + +/* Provider interface to dpifs, which provide an interface to an Open vSwitch + * datapath. */ + +#include +#include "dpif.h" + +/* Open vSwitch datapath interface. + * + * This structure should be treated as opaque by dpif implementations. */ +struct dpif { + const struct dpif_class *class; + char *name; + uint8_t netflow_engine_type; + uint8_t netflow_engine_id; +}; + +void dpif_init(struct dpif *, const struct dpif_class *, const char *name, + uint8_t netflow_engine_type, uint8_t netflow_engine_id); +static inline void dpif_assert_class(const struct dpif *dpif, + const struct dpif_class *class) +{ + assert(dpif->class == class); +} + +/* Datapath interface class structure, to be defined by each implementation of + * a datapath interface + * + * These functions return 0 if successful or a positive errno value on failure, + * except where otherwise noted. + * + * These functions are expected to execute synchronously, that is, to block as + * necessary to obtain a result. Thus, they may not return EAGAIN or + * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if + * and when we encounter performance problems. */ +struct dpif_class { + /* Prefix for names of dpifs in this class, e.g. "udatapath:". + * + * One dpif class may have the empty string "" as its prefix, in which case + * that dpif class is associated with dpif names that don't match any other + * class name. */ + const char *prefix; + + /* Class name, for use in error messages. */ + const char *name; + + /* Attempts to open an existing dpif, if 'create' is false, or to open an + * existing dpif or create a new one, if 'create' is true. 'name' is the + * full dpif name provided by the user, e.g. "udatapath:/var/run/mypath". + * This name is useful for error messages but must not be modified. + * + * 'suffix' is a copy of 'name' following the dpif's 'prefix'. + * + * If successful, stores a pointer to the new dpif in '*dpifp'. On failure + * there are no requirements on what is stored in '*dpifp'. */ + int (*open)(const char *name, char *suffix, bool create, + struct dpif **dpifp); + + /* Closes 'dpif' and frees associated memory. */ + void (*close)(struct dpif *dpif); + + /* Attempts to destroy the dpif underlying 'dpif'. + * + * If successful, 'dpif' will not be used again except as an argument for + * the 'close' member function. */ + int (*delete)(struct dpif *dpif); + + /* Retrieves statistics for 'dpif' into 'stats'. */ + int (*get_stats)(const struct dpif *dpif, struct odp_stats *stats); + + /* Retrieves 'dpif''s current treatment of IP fragments into '*drop_frags': + * true indicates that fragments are dropped, false indicates that + * fragments are treated in the same way as other IP packets (except that + * the L4 header cannot be read). */ + int (*get_drop_frags)(const struct dpif *dpif, bool *drop_frags); + + /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose + * meaning is the same as for the get_drop_frags member function. */ + int (*set_drop_frags)(struct dpif *dpif, bool drop_frags); + + /* Creates a new port in 'dpif' connected to network device 'devname'. + * 'flags' is a set of ODP_PORT_* flags. If successful, sets '*port_no' + * to the new port's port number. */ + int (*port_add)(struct dpif *dpif, const char *devname, uint16_t flags, + uint16_t *port_no); + + /* Removes port numbered 'port_no' from 'dpif'. */ + int (*port_del)(struct dpif *dpif, uint16_t port_no); + + /* Queries 'dpif' for a port with the given 'port_no' or 'devname'. Stores + * information about the port into '*port' if successful. */ + int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no, + struct odp_port *port); + int (*port_query_by_name)(const struct dpif *dpif, const char *devname, + struct odp_port *port); + + /* Stores in 'ports' information about up to 'n' ports attached to 'dpif', + * in no particular order. Returns the number of ports attached to 'dpif' + * (not the number stored), if successful, otherwise a negative errno + * value. */ + int (*port_list)(const struct dpif *dpif, struct odp_port *ports, int n); + + /* Stores in 'ports' the port numbers of up to 'n' ports that belong to + * 'group' in 'dpif'. Returns the number of ports in 'group' (not the + * number stored), if successful, otherwise a negative errno value. */ + int (*port_group_get)(const struct dpif *dpif, int group, + uint16_t ports[], int n); + + /* Changes port group 'group' in 'dpif' to consist of the 'n' ports whose + * numbers are given in 'ports'. + * + * Use the get_stats member function to obtain the number of supported port + * groups. */ + int (*port_group_set)(struct dpif *dpif, int group, + const uint16_t ports[], int n); + + /* For each flow 'flow' in the 'n' flows in 'flows': + * + * - If a flow matching 'flow->key' exists in 'dpif': + * + * Stores 0 into 'flow->stats.error' and stores statistics for the flow + * into 'flow->stats'. + * + * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If + * 'flow->n_actions' is nonzero, then 'flow->actions' should point to + * an array of the specified number of actions. At most that many of + * the flow's actions will be copied into that array. + * 'flow->n_actions' will be updated to the number of actions actually + * present in the flow, which may be greater than the number stored if + * the flow has more actions than space available in the array. + * + * - Flow-specific errors are indicated by a positive errno value in + * 'flow->stats.error'. In particular, ENOENT indicates that no flow + * matching 'flow->key' exists in 'dpif'. When an error value is stored, + * the contents of 'flow->key' are preserved but other members of 'flow' + * should be treated as indeterminate. + * + * Returns 0 if all 'n' flows in 'flows' were updated (whether they were + * individually successful or not is indicated by 'flow->stats.error', + * however). Returns a positive errno value if an error that prevented + * this update occurred, in which the caller must not depend on any + * elements in 'flows' being updated or not updated. + */ + int (*flow_get)(const struct dpif *dpif, struct odp_flow flows[], int n); + + /* Adds or modifies a flow in 'dpif' as specified in 'put': + * + * - If the flow specified in 'put->flow' does not exist in 'dpif', then + * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': + * if it is, the flow will be added, otherwise the operation will fail + * with ENOENT. + * + * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'. + * Behavior in this case depends on whether ODPPF_MODIFY is specified in + * 'put->flags': if it is, the flow's actions will be updated, otherwise + * the operation will fail with EEXIST. If the flow's actions are + * updated, then its statistics will be zeroed if ODPPF_ZERO_STATS is set + * in 'put->flags', left as-is otherwise. + */ + int (*flow_put)(struct dpif *dpif, struct odp_flow_put *put); + + /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if + * 'dpif' does not contain such a flow. + * + * If successful, updates 'flow->stats', 'flow->n_actions', and + * 'flow->actions' as described in more detail under the flow_get member + * function below. */ + int (*flow_del)(struct dpif *dpif, struct odp_flow *flow); + + /* Deletes all flows from 'dpif' and clears all of its queues of received + * packets. */ + int (*flow_flush)(struct dpif *dpif); + + /* Stores up to 'n' flows in 'dpif' into 'flows', updating their statistics + * and actions as described under the flow_get member function. If + * successful, returns the number of flows actually present in 'dpif', + * which might be greater than the number stored (if 'dpif' has more than + * 'n' flows). On failure, returns a negative errno value. */ + int (*flow_list)(const struct dpif *dpif, struct odp_flow flows[], int n); + + /* Performs the 'n_actions' actions in 'actions' on the Ethernet frame + * specified in 'packet'. + * + * Pretends that the frame was originally received on the port numbered + * 'in_port'. This affects only ODPAT_OUTPUT_GROUP actions, which will not + * send a packet out their input port. Specify the number of an unused + * port (e.g. UINT16_MAX is currently always unused) to avoid this + * behavior. */ + int (*execute)(struct dpif *dpif, uint16_t in_port, + const union odp_action actions[], int n_actions, + const struct ofpbuf *packet); + + /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit + * set in '*listen_mask' indicates the 'dpif' will receive messages of the + * corresponding type when it calls the recv member function. */ + int (*recv_get_mask)(const struct dpif *dpif, int *listen_mask); + + /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in + * 'listen_mask' indicates the 'dpif' will receive messages of the + * corresponding type when it calls the recv member function. */ + int (*recv_set_mask)(struct dpif *dpif, int listen_mask); + + /* Attempts to receive a message from 'dpif'. If successful, stores the + * message into '*packetp'. The message, if one is received, must begin + * with 'struct odp_msg' as a header. Only messages of the types selected + * with the set_listen_mask member function should be received. + * + * This function must not block. If no message is ready to be received + * when it is called, it should return EAGAIN without blocking. */ + int (*recv)(struct dpif *dpif, struct ofpbuf **packetp); + + /* Arranges for the poll loop to wake up when 'dpif' has a message queued + * to be received with the recv member function. */ + void (*recv_wait)(struct dpif *dpif); +}; + +extern const struct dpif_class dpif_linux_class; + +#endif /* dpif-provider.h */ diff --git a/lib/dpif.c b/lib/dpif.c index 85d05bb7..4159dbd0 100644 --- a/lib/dpif.c +++ b/lib/dpif.c @@ -15,24 +15,14 @@ */ #include -#include "dpif.h" +#include "dpif-provider.h" #include #include #include -#include #include -#include -#include -#include -#include -#include #include #include -#include -#include -#include -#include #include "coverage.h" #include "dynamic-string.h" @@ -49,12 +39,10 @@ #include "vlog.h" #define THIS_MODULE VLM_dpif -/* A datapath interface. */ -struct dpif { - char *name; - unsigned int minor; - int fd; +static struct dpif_class *dpif_classes[] = { + &dpif_linux_class, }; +enum { N_DPIF_CLASSES = ARRAY_SIZE(dpif_classes) }; /* Rate limit for individual messages going to or from the datapath, output at * DBG level. This is very high because, if these are enabled, it is because @@ -64,207 +52,164 @@ static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600); /* Not really much point in logging many dpif errors. */ static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); -static int get_minor_from_name(const char *name, unsigned int *minor); -static int name_to_minor(const char *name, unsigned int *minor); -static int lookup_minor(const char *name, unsigned int *minor); -static int open_by_minor(unsigned int minor, struct dpif **dpifp); -static int make_openvswitch_device(unsigned int minor, char **fnp); +static void log_operation(const struct dpif *, const char *operation, + int error); +static void log_flow_operation(const struct dpif *, const char *operation, + int error, struct odp_flow *flow); +static void log_flow_put(struct dpif *, int error, + const struct odp_flow_put *); +static bool should_log_flow_message(int error); static void check_rw_odp_flow(struct odp_flow *); -int -dpif_open(const char *name, struct dpif **dpifp) +static int +do_open(const char *name_, bool create, struct dpif **dpifp) { - struct dpif *dpif; - unsigned int minor; - int listen_mask; + char *name = xstrdup(name_); + char *prefix, *suffix, *colon; + struct dpif *dpif = NULL; int error; + int i; - *dpifp = NULL; - - error = name_to_minor(name, &minor); - if (error) { - return error; - } - - error = open_by_minor(minor, &dpif); - if (error) { - return error; + colon = strchr(name, ':'); + if (colon) { + *colon = '\0'; + prefix = name; + suffix = colon + 1; + } else { + prefix = ""; + suffix = name; } - /* We can open the device, but that doesn't mean that it's been created. - * If it hasn't been, then any command other than ODP_DP_CREATE will - * return ENODEV. Try something innocuous. */ - listen_mask = 0; /* Make Valgrind happy. */ - if (ioctl(dpif->fd, ODP_GET_LISTEN_MASK, &listen_mask)) { - error = errno; - if (error != ENODEV) { - VLOG_WARN("%s: probe returned unexpected error: %s", - dpif_name(dpif), strerror(error)); + for (i = 0; i < N_DPIF_CLASSES; i++) { + const struct dpif_class *class = dpif_classes[i]; + if (!strcmp(prefix, class->prefix)) { + error = class->open(name_, suffix, create, &dpif); + goto exit; } - dpif_close(dpif); - return error; } - *dpifp = dpif; - return 0; -} + error = EAFNOSUPPORT; -void -dpif_close(struct dpif *dpif) -{ - if (dpif) { - free(dpif->name); - close(dpif->fd); - free(dpif); - } +exit: + *dpifp = error ? NULL : dpif; + return error; } -static int -do_ioctl(const struct dpif *dpif, int cmd, const char *cmd_name, - const void *arg) +/* Tries to open an existing datapath named 'name'. Will fail if no datapath + * named 'name' exists. Returns 0 if successful, otherwise a positive errno + * value. On success stores a pointer to the datapath in '*dpifp', otherwise a + * null pointer. */ +int +dpif_open(const char *name, struct dpif **dpifp) { - int error = ioctl(dpif->fd, cmd, arg) ? errno : 0; - if (cmd_name) { - if (error) { - VLOG_WARN_RL(&error_rl, "%s: ioctl(%s) failed (%s)", - dpif_name(dpif), cmd_name, strerror(error)); - } else { - VLOG_DBG_RL(&dpmsg_rl, "%s: ioctl(%s): success", - dpif_name(dpif), cmd_name); - } - } - return error; + return do_open(name, false, dpifp); } +/* Tries to create and open a new datapath with the given 'name'. Will fail if + * a datapath named 'name' already exists. Returns 0 if successful, otherwise + * a positive errno value. On success stores a pointer to the datapath in + * '*dpifp', otherwise a null pointer.*/ int dpif_create(const char *name, struct dpif **dpifp) { - unsigned int minor; - int error; - - *dpifp = NULL; - if (!get_minor_from_name(name, &minor)) { - /* Minor was specified in 'name', go ahead and create it. */ - struct dpif *dpif; - - error = open_by_minor(minor, &dpif); - if (error) { - return error; - } - - error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0; - if (!error) { - *dpifp = dpif; - } else { - dpif_close(dpif); - } - return error; - } else { - for (minor = 0; minor < ODP_MAX; minor++) { - struct dpif *dpif; - - error = open_by_minor(minor, &dpif); - if (error) { - return error; - } + return do_open(name, true, dpifp); +} - error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0; - if (!error) { - *dpifp = dpif; - return 0; - } - dpif_close(dpif); - if (error != EBUSY) { - return error; - } - } - return ENOBUFS; +/* Closes and frees the connection to 'dpif'. Does not destroy the datapath + * itself; call dpif_delete() first, instead, if that is desirable. */ +void +dpif_close(struct dpif *dpif) +{ + if (dpif) { + char *name = dpif->name; + dpif->class->close(dpif); + free(name); } } +/* Returns the name of datapath 'dpif' (for use in log messages). */ const char * dpif_name(const struct dpif *dpif) { return dpif->name; } +/* Destroys the datapath that 'dpif' is connected to, first removing all of its + * ports. After calling this function, it does not make sense to pass 'dpif' + * to any functions other than dpif_name() or dpif_close(). */ int dpif_delete(struct dpif *dpif) { + int error; + COVERAGE_INC(dpif_destroy); - return do_ioctl(dpif, ODP_DP_DESTROY, "ODP_DP_DESTROY", NULL); + + error = dpif->class->delete(dpif); + log_operation(dpif, "delete", error); + return error; } +/* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful, + * otherwise a positive errno value. */ int dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats) { - memset(stats, 0, sizeof *stats); - return do_ioctl(dpif, ODP_DP_STATS, "ODP_DP_STATS", stats); + int error = dpif->class->get_stats(dpif, stats); + if (error) { + memset(stats, 0, sizeof *stats); + } + log_operation(dpif, "get_stats", error); + return error; } +/* Retrieves the current IP fragment handling policy for 'dpif' into + * '*drop_frags': true indicates that fragments are dropped, false indicates + * that fragments are treated in the same way as other IP packets (except that + * the L4 header cannot be read). Returns 0 if successful, otherwise a + * positive errno value. */ int dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags) { - int tmp; - int error = do_ioctl(dpif, ODP_GET_DROP_FRAGS, "ODP_GET_DROP_FRAGS", &tmp); - *drop_frags = error ? tmp & 1 : false; + int error = dpif->class->get_drop_frags(dpif, drop_frags); + if (error) { + *drop_frags = false; + } + log_operation(dpif, "get_drop_frags", error); return error; } +/* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is + * the same as for the get_drop_frags member function. Returns 0 if + * successful, otherwise a positive errno value. */ int dpif_set_drop_frags(struct dpif *dpif, bool drop_frags) { - int tmp = drop_frags; - return do_ioctl(dpif, ODP_SET_DROP_FRAGS, "ODP_SET_DROP_FRAGS", &tmp); -} - -int -dpif_recv_purge(struct dpif *dpif) -{ - struct odp_stats stats; - unsigned int i; - int error; - - COVERAGE_INC(dpif_purge); - - error = dpif_get_dp_stats(dpif, &stats); - if (error) { - return error; - } - - for (i = 0; i < stats.max_miss_queue + stats.max_action_queue; i++) { - struct ofpbuf *buf; - error = dpif_recv(dpif, &buf); - if (error) { - return error == EAGAIN ? 0 : error; - } - ofpbuf_delete(buf); - } - return 0; + int error = dpif->class->set_drop_frags(dpif, drop_frags); + log_operation(dpif, "set_drop_frags", error); + return error; } +/* Attempts to add 'devname' as a port on 'dpif', given the combination of + * ODP_PORT_* flags in 'flags'. If successful, returns 0 and sets '*port_nop' + * to the new port's port number (if 'port_nop' is non-null). On failure, + * returns a positive errno value and sets '*port_nop' to UINT16_MAX (if + * 'port_nop' is non-null). */ int dpif_port_add(struct dpif *dpif, const char *devname, uint16_t flags, uint16_t *port_nop) { - struct odp_port port; uint16_t port_no; int error; COVERAGE_INC(dpif_port_add); - memset(&port, 0, sizeof port); - strncpy(port.devname, devname, sizeof port.devname); - port.flags = flags; - - error = do_ioctl(dpif, ODP_PORT_ADD, NULL, &port); + error = dpif->class->port_add(dpif, devname, flags, &port_no); if (!error) { - port_no = port.port; VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16, dpif_name(dpif), devname, port_no); } else { - port_no = UINT16_MAX; VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s", - dpif_name(dpif), devname, strerror(errno)); + dpif_name(dpif), devname, strerror(error)); + port_no = UINT16_MAX; } if (port_nop) { *port_nop = port_no; @@ -272,51 +217,66 @@ dpif_port_add(struct dpif *dpif, const char *devname, uint16_t flags, return error; } +/* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful, + * otherwise a positive errno value. */ int dpif_port_del(struct dpif *dpif, uint16_t port_no) { - int tmp = port_no; + int error; + COVERAGE_INC(dpif_port_del); - return do_ioctl(dpif, ODP_PORT_DEL, "ODP_PORT_DEL", &tmp); + + error = dpif->class->port_del(dpif, port_no); + log_operation(dpif, "port_del", error); + return error; } +/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and + * initializes '*port' appropriately; on failure, returns a positive errno + * value. */ int dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no, struct odp_port *port) { - memset(port, 0, sizeof *port); - port->port = port_no; - if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) { + int error = dpif->class->port_query_by_number(dpif, port_no, port); + if (!error) { VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s", dpif_name(dpif), port_no, port->devname); - return 0; } else { + memset(port, 0, sizeof *port); VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s", - dpif_name(dpif), port_no, strerror(errno)); - return errno; + dpif_name(dpif), port_no, strerror(error)); } + return error; } +/* Looks up port named 'devname' in 'dpif'. On success, returns 0 and + * initializes '*port' appropriately; on failure, returns a positive errno + * value. */ int dpif_port_query_by_name(const struct dpif *dpif, const char *devname, struct odp_port *port) { - memset(port, 0, sizeof *port); - strncpy(port->devname, devname, sizeof port->devname); - if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) { + int error = dpif->class->port_query_by_name(dpif, devname, port); + if (!error) { VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16, dpif_name(dpif), devname, port->port); - return 0; } else { + memset(port, 0, sizeof *port); + /* Log level is DBG here because all the current callers are interested * in whether 'dpif' actually has a port 'devname', so that it's not an * issue worth logging if it doesn't. */ VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s", - dpif_name(dpif), devname, strerror(errno)); - return errno; + dpif_name(dpif), devname, strerror(error)); } + return error; } +/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies + * the port's name into the 'name_size' bytes in 'name', ensuring that the + * result is null-terminated. On failure, returns a positive errno value and + * makes 'name' the empty string. */ int dpif_port_get_name(struct dpif *dpif, uint16_t port_no, char *name, size_t name_size) @@ -335,6 +295,15 @@ dpif_port_get_name(struct dpif *dpif, uint16_t port_no, return error; } +/* Obtains a list of all the ports in 'dpif'. + * + * If successful, returns 0 and sets '*portsp' to point to an array of + * appropriately initialized port structures and '*n_portsp' to the number of + * ports in the array. The caller is responsible for freeing '*portp' by + * calling free(). + * + * On failure, returns a positive errno value and sets '*portsp' to NULL and + * '*n_portsp' to 0. */ int dpif_port_list(const struct dpif *dpif, struct odp_port **portsp, size_t *n_portsp) @@ -345,7 +314,7 @@ dpif_port_list(const struct dpif *dpif, for (;;) { struct odp_stats stats; - struct odp_portvec pv; + int retval; error = dpif_get_dp_stats(dpif, &stats); if (error) { @@ -353,17 +322,16 @@ dpif_port_list(const struct dpif *dpif, } ports = xcalloc(stats.n_ports, sizeof *ports); - pv.ports = ports; - pv.n_ports = stats.n_ports; - error = do_ioctl(dpif, ODP_PORT_LIST, "ODP_PORT_LIST", &pv); - if (error) { + retval = dpif->class->port_list(dpif, ports, stats.n_ports); + if (retval < 0) { /* Hard error. */ + error = -retval; free(ports); goto exit; - } else if (pv.n_ports <= stats.n_ports) { + } else if (retval <= stats.n_ports) { /* Success. */ error = 0; - n_ports = pv.n_ports; + n_ports = retval; goto exit; } else { /* Soft error: port count increased behind our back. Try again. */ @@ -379,23 +347,19 @@ exit: *portsp = ports; *n_portsp = n_ports; } + log_operation(dpif, "port_list", error); return error; } -int -dpif_port_group_set(struct dpif *dpif, uint16_t group, - const uint16_t ports[], size_t n_ports) -{ - struct odp_port_group pg; - - COVERAGE_INC(dpif_port_group_set); - assert(n_ports <= UINT16_MAX); - pg.group = group; - pg.ports = (uint16_t *) ports; - pg.n_ports = n_ports; - return do_ioctl(dpif, ODP_PORT_GROUP_SET, "ODP_PORT_GROUP_SET", &pg); -} - +/* Retrieves a list of the port numbers in port group 'group' in 'dpif'. + * + * On success, returns 0 and points '*ports' to a newly allocated array of + * integers, each of which is a 'dpif' port number for a port in + * 'group'. Stores the number of elements in the array in '*n_ports'. The + * caller is responsible for freeing '*ports' by calling free(). + * + * On failure, returns a positive errno value and sets '*ports' to NULL and + * '*n_ports' to 0. */ int dpif_port_group_get(const struct dpif *dpif, uint16_t group, uint16_t **ports, size_t *n_ports) @@ -405,185 +369,206 @@ dpif_port_group_get(const struct dpif *dpif, uint16_t group, *ports = NULL; *n_ports = 0; for (;;) { - struct odp_port_group pg; - pg.group = group; - pg.ports = *ports; - pg.n_ports = *n_ports; - - error = do_ioctl(dpif, ODP_PORT_GROUP_GET, "ODP_PORT_GROUP_GET", &pg); - if (error) { + int retval = dpif->class->port_group_get(dpif, group, + *ports, *n_ports); + if (retval < 0) { /* Hard error. */ + error = -retval; free(*ports); *ports = NULL; *n_ports = 0; break; - } else if (pg.n_ports <= *n_ports) { + } else if (retval <= *n_ports) { /* Success. */ - *n_ports = pg.n_ports; + error = 0; + *n_ports = retval; break; } else { /* Soft error: there were more ports than we expected in the * group. Try again. */ free(*ports); - *ports = xcalloc(pg.n_ports, sizeof **ports); - *n_ports = pg.n_ports; + *ports = xcalloc(retval, sizeof **ports); + *n_ports = retval; } } + log_operation(dpif, "port_group_get", error); return error; } +/* Updates port group 'group' in 'dpif', making it contain the 'n_ports' ports + * whose 'dpif' port numbers are given in 'n_ports'. Returns 0 if + * successful, otherwise a positive errno value. + * + * Behavior is undefined if the values in ports[] are not unique. */ int -dpif_flow_flush(struct dpif *dpif) +dpif_port_group_set(struct dpif *dpif, uint16_t group, + const uint16_t ports[], size_t n_ports) { - COVERAGE_INC(dpif_flow_flush); - return do_ioctl(dpif, ODP_FLOW_FLUSH, "ODP_FLOW_FLUSH", NULL); -} + int error; -static enum vlog_level -flow_message_log_level(int error) -{ - return error ? VLL_WARN : VLL_DBG; + COVERAGE_INC(dpif_port_group_set); + + error = dpif->class->port_group_set(dpif, group, ports, n_ports); + log_operation(dpif, "port_group_set", error); + return error; } -static bool -should_log_flow_message(int error) +/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a + * positive errno value. */ +int +dpif_flow_flush(struct dpif *dpif) { - return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error), - error ? &error_rl : &dpmsg_rl); + int error; + + COVERAGE_INC(dpif_flow_flush); + + error = dpif->class->flow_flush(dpif); + log_operation(dpif, "flow_flush", error); + return error; } -static void -log_flow_message(const struct dpif *dpif, int error, - const char *operation, - const flow_t *flow, const struct odp_flow_stats *stats, - const union odp_action *actions, size_t n_actions) +/* Queries 'dpif' for a flow entry matching 'flow->key'. + * + * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the + * flow into 'flow->stats'. If 'flow->n_actions' is zero, then 'flow->actions' + * is ignored. If 'flow->n_actions' is nonzero, then 'flow->actions' should + * point to an array of the specified number of actions. At most that many of + * the flow's actions will be copied into that array. 'flow->n_actions' will + * be updated to the number of actions actually present in the flow, which may + * be greater than the number stored if the flow has more actions than space + * available in the array. + * + * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other + * failure, returns a positive errno value. */ +int +dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow) { - struct ds ds = DS_EMPTY_INITIALIZER; - ds_put_format(&ds, "%s: ", dpif_name(dpif)); - if (error) { - ds_put_cstr(&ds, "failed to "); - } - ds_put_format(&ds, "%s ", operation); - if (error) { - ds_put_format(&ds, "(%s) ", strerror(error)); - } - flow_format(&ds, flow); - if (stats) { - ds_put_cstr(&ds, ", "); - format_odp_flow_stats(&ds, stats); + int error; + + COVERAGE_INC(dpif_flow_get); + + check_rw_odp_flow(flow); + error = dpif->class->flow_get(dpif, flow, 1); + if (!error) { + error = flow->stats.error; } - if (actions || n_actions) { - ds_put_cstr(&ds, ", actions:"); - format_odp_actions(&ds, actions, n_actions); + if (should_log_flow_message(error)) { + log_flow_operation(dpif, "flow_get", error, flow); } - vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds)); - ds_destroy(&ds); + return error; } -static int -do_flow_ioctl(const struct dpif *dpif, int cmd, struct odp_flow *flow, - const char *operation, bool show_stats) +/* For each flow 'flow' in the 'n' flows in 'flows': + * + * - If a flow matching 'flow->key' exists in 'dpif': + * + * Stores 0 into 'flow->stats.error' and stores statistics for the flow + * into 'flow->stats'. + * + * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If + * 'flow->n_actions' is nonzero, then 'flow->actions' should point to an + * array of the specified number of actions. At most that many of the + * flow's actions will be copied into that array. 'flow->n_actions' will + * be updated to the number of actions actually present in the flow, which + * may be greater than the number stored if the flow has more actions than + * space available in the array. + * + * - Flow-specific errors are indicated by a positive errno value in + * 'flow->stats.error'. In particular, ENOENT indicates that no flow + * matching 'flow->key' exists in 'dpif'. When an error value is stored, the + * contents of 'flow->key' are preserved but other members of 'flow' should + * be treated as indeterminate. + * + * Returns 0 if all 'n' flows in 'flows' were updated (whether they were + * individually successful or not is indicated by 'flow->stats.error', + * however). Returns a positive errno value if an error that prevented this + * update occurred, in which the caller must not depend on any elements in + * 'flows' being updated or not updated. + */ +int +dpif_flow_get_multiple(const struct dpif *dpif, + struct odp_flow flows[], size_t n) { - int error = do_ioctl(dpif, cmd, NULL, flow); - if (error && show_stats) { - flow->n_actions = 0; - } - if (should_log_flow_message(error)) { - log_flow_message(dpif, error, operation, &flow->key, - show_stats && !error ? &flow->stats : NULL, - flow->actions, flow->n_actions); + int error; + size_t i; + + COVERAGE_ADD(dpif_flow_get, n); + + for (i = 0; i < n; i++) { + check_rw_odp_flow(&flows[i]); } + + error = dpif->class->flow_get(dpif, flows, n); + log_operation(dpif, "flow_get_multiple", error); return error; } +/* Adds or modifies a flow in 'dpif' as specified in 'put': + * + * - If the flow specified in 'put->flow' does not exist in 'dpif', then + * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if + * it is, the flow will be added, otherwise the operation will fail with + * ENOENT. + * + * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'. + * Behavior in this case depends on whether ODPPF_MODIFY is specified in + * 'put->flags': if it is, the flow's actions will be updated, otherwise the + * operation will fail with EEXIST. If the flow's actions are updated, then + * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags', + * left as-is otherwise. + * + * Returns 0 if successful, otherwise a positive errno value. + */ int dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put) { - int error = do_ioctl(dpif, ODP_FLOW_PUT, NULL, put); + int error; + COVERAGE_INC(dpif_flow_put); + + error = dpif->class->flow_put(dpif, put); if (should_log_flow_message(error)) { - struct ds operation = DS_EMPTY_INITIALIZER; - ds_put_cstr(&operation, "put"); - if (put->flags & ODPPF_CREATE) { - ds_put_cstr(&operation, "[create]"); - } - if (put->flags & ODPPF_MODIFY) { - ds_put_cstr(&operation, "[modify]"); - } - if (put->flags & ODPPF_ZERO_STATS) { - ds_put_cstr(&operation, "[zero]"); - } -#define ODPPF_ALL (ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS) - if (put->flags & ~ODPPF_ALL) { - ds_put_format(&operation, "[%x]", put->flags & ~ODPPF_ALL); - } - log_flow_message(dpif, error, ds_cstr(&operation), &put->flow.key, - !error ? &put->flow.stats : NULL, - put->flow.actions, put->flow.n_actions); - ds_destroy(&operation); + log_flow_put(dpif, error, put); } return error; } +/* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif' + * does not contain such a flow. + * + * If successful, updates 'flow->stats', 'flow->n_actions', and 'flow->actions' + * as described for dpif_flow_get(). */ int dpif_flow_del(struct dpif *dpif, struct odp_flow *flow) { - COVERAGE_INC(dpif_flow_del); - check_rw_odp_flow(flow); - memset(&flow->stats, 0, sizeof flow->stats); - return do_flow_ioctl(dpif, ODP_FLOW_DEL, flow, "delete flow", true); -} - -int -dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow) -{ - struct odp_flowvec fv; int error; - COVERAGE_INC(dpif_flow_get); + COVERAGE_INC(dpif_flow_del); check_rw_odp_flow(flow); - fv.flows = flow; - fv.n_flows = 1; - error = do_ioctl(dpif, ODP_FLOW_GET, "ODP_FLOW_GET", &fv); - if (!error) { - error = flow->stats.error; - if (error) { - VLOG_WARN_RL(&error_rl, "%s: ioctl(ODP_FLOW_GET) failed (%s)", - dpif_name(dpif), strerror(error)); - } - } - return error; -} - -int -dpif_flow_get_multiple(const struct dpif *dpif, - struct odp_flow flows[], size_t n) -{ - struct odp_flowvec fv; - size_t i; + memset(&flow->stats, 0, sizeof flow->stats); - COVERAGE_ADD(dpif_flow_query_multiple, n); - fv.flows = flows; - fv.n_flows = n; - for (i = 0; i < n; i++) { - check_rw_odp_flow(&flows[i]); + error = dpif->class->flow_del(dpif, flow); + if (should_log_flow_message(error)) { + log_flow_operation(dpif, "delete flow", error, flow); } - return do_ioctl(dpif, ODP_FLOW_GET, "ODP_FLOW_GET", - &fv); + return error; } +/* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics + * but not including any information about their actions. If successful, + * returns 0 and sets '*n_out' to the number of flows actually present in + * 'dpif', which might be greater than the number stored (if 'dpif' has more + * than 'n' flows). On failure, returns a negative errno value and sets + * '*n_out' to 0. */ int dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n, size_t *n_out) { - struct odp_flowvec fv; uint32_t i; - int error; + int retval; COVERAGE_INC(dpif_flow_query_list); - fv.flows = flows; - fv.n_flows = n; if (RUNNING_ON_VALGRIND) { memset(flows, 0, n * sizeof *flows); } else { @@ -592,20 +577,31 @@ dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n, flows[i].n_actions = 0; } } - error = do_ioctl(dpif, ODP_FLOW_LIST, NULL, &fv); - if (error) { + retval = dpif->class->flow_list(dpif, flows, n); + if (retval < 0) { *n_out = 0; VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)", - dpif_name(dpif), strerror(error)); + dpif_name(dpif), strerror(-retval)); + return -retval; } else { - COVERAGE_ADD(dpif_flow_query_list_n, fv.n_flows); - *n_out = fv.n_flows; - VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows", - dpif_name(dpif), *n_out); + COVERAGE_ADD(dpif_flow_query_list_n, retval); + *n_out = MIN(n, retval); + VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows (of %d)", + dpif_name(dpif), *n_out, retval); + return 0; } - return error; } +/* Retrieves all of the flows in 'dpif'. + * + * If successful, returns 0 and stores in '*flowsp' a pointer to a newly + * allocated array of flows, including their statistics but not including any + * information about their actions, and sets '*np' to the number of flows in + * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling + * free(). + * + * On failure, returns a positive errno value and sets '*flowsp' to NULL and + * '*np' to 0. */ int dpif_flow_list_all(const struct dpif *dpif, struct odp_flow **flowsp, size_t *np) @@ -640,6 +636,15 @@ dpif_flow_list_all(const struct dpif *dpif, return 0; } +/* Causes 'dpif' to perform the 'n_actions' actions in 'actions' on the + * Ethernet frame specified in 'packet'. + * + * Pretends that the frame was originally received on the port numbered + * 'in_port'. This affects only ODPAT_OUTPUT_GROUP actions, which will not + * send a packet out their input port. Specify the number of an unused port + * (e.g. UINT16_MAX is currently always unused) to avoid this behavior. + * + * Returns 0 if successful, otherwise a positive errno value. */ int dpif_execute(struct dpif *dpif, uint16_t in_port, const union odp_action actions[], size_t n_actions, @@ -649,14 +654,7 @@ dpif_execute(struct dpif *dpif, uint16_t in_port, COVERAGE_INC(dpif_execute); if (n_actions > 0) { - struct odp_execute execute; - memset(&execute, 0, sizeof execute); - execute.in_port = in_port; - execute.actions = (union odp_action *) actions; - execute.n_actions = n_actions; - execute.data = buf->data; - execute.length = buf->size; - error = do_ioctl(dpif, ODP_EXECUTE, NULL, &execute); + error = dpif->class->execute(dpif, in_port, actions, n_actions, buf); } else { error = 0; } @@ -677,93 +675,242 @@ dpif_execute(struct dpif *dpif, uint16_t in_port, return error; } +/* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set + * in '*listen_mask' indicates that dpif_recv() will receive messages of that + * type. Returns 0 if successful, otherwise a positive errno value. */ int dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask) { - int error = do_ioctl(dpif, ODP_GET_LISTEN_MASK, "ODP_GET_LISTEN_MASK", - listen_mask); + int error = dpif->class->recv_get_mask(dpif, listen_mask); if (error) { *listen_mask = 0; } + log_operation(dpif, "recv_get_mask", error); return error; } +/* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in + * '*listen_mask' requests that dpif_recv() receive messages of that type. + * Returns 0 if successful, otherwise a positive errno value. */ int dpif_recv_set_mask(struct dpif *dpif, int listen_mask) { - return do_ioctl(dpif, ODP_SET_LISTEN_MASK, "ODP_SET_LISTEN_MASK", - &listen_mask); + int error = dpif->class->recv_set_mask(dpif, listen_mask); + log_operation(dpif, "recv_set_mask", error); + return error; } +/* Attempts to receive a message from 'dpif'. If successful, stores the + * message into '*packetp'. The message, if one is received, will begin with + * 'struct odp_msg' as a header. Only messages of the types selected with + * dpif_set_listen_mask() will ordinarily be received (but if a message type is + * enabled and then later disabled, some stragglers might pop up). + * + * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN + * if no message is immediately available. */ int -dpif_recv(struct dpif *dpif, struct ofpbuf **bufp) +dpif_recv(struct dpif *dpif, struct ofpbuf **packetp) { - struct ofpbuf *buf; - int retval; - int error; - - buf = ofpbuf_new(65536); - retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); - if (retval < 0) { - error = errno; - if (error != EAGAIN) { - VLOG_WARN_RL(&error_rl, "%s: read failed: %s", - dpif_name(dpif), strerror(error)); - } - } else if (retval >= sizeof(struct odp_msg)) { - struct odp_msg *msg = buf->data; - if (msg->length <= retval) { - buf->size += retval; - if (VLOG_IS_DBG_ENABLED()) { - void *payload = msg + 1; - size_t length = buf->size - sizeof *msg; - char *s = ofp_packet_to_string(payload, length, length); - VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length " - "%zu on port %"PRIu16": %s", dpif_name(dpif), - (msg->type == _ODPL_MISS_NR ? "miss" - : msg->type == _ODPL_ACTION_NR ? "action" - : ""), - msg->length - sizeof(struct odp_msg), - msg->port, s); - free(s); - } - *bufp = buf; - COVERAGE_INC(dpif_recv); - return 0; - } else { - VLOG_WARN_RL(&error_rl, "%s: discarding message truncated " - "from %zu bytes to %d", - dpif_name(dpif), msg->length, retval); - error = ERANGE; + int error = dpif->class->recv(dpif, packetp); + if (!error) { + if (VLOG_IS_DBG_ENABLED()) { + struct ofpbuf *buf = *packetp; + struct odp_msg *msg = buf->data; + void *payload = msg + 1; + size_t payload_len = buf->size - sizeof *msg; + char *s = ofp_packet_to_string(payload, payload_len, payload_len); + VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length " + "%zu on port %"PRIu16": %s", dpif_name(dpif), + (msg->type == _ODPL_MISS_NR ? "miss" + : msg->type == _ODPL_ACTION_NR ? "action" + : ""), + payload_len, msg->port, s); + free(s); } - } else if (!retval) { - VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif)); - error = EPROTO; } else { - VLOG_WARN_RL(&error_rl, - "%s: discarding too-short message (%d bytes)", - dpif_name(dpif), retval); - error = ERANGE; + *packetp = NULL; } - - *bufp = NULL; - ofpbuf_delete(buf); return error; } +/* Discards all messages that would otherwise be received by dpif_recv() on + * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */ +int +dpif_recv_purge(struct dpif *dpif) +{ + struct odp_stats stats; + unsigned int i; + int error; + + COVERAGE_INC(dpif_purge); + + error = dpif_get_dp_stats(dpif, &stats); + if (error) { + return error; + } + + for (i = 0; i < stats.max_miss_queue + stats.max_action_queue; i++) { + struct ofpbuf *buf; + error = dpif_recv(dpif, &buf); + if (error) { + return error == EAGAIN ? 0 : error; + } + ofpbuf_delete(buf); + } + return 0; +} + +/* Arranges for the poll loop to wake up when 'dpif' has a message queued to be + * received with dpif_recv(). */ void dpif_recv_wait(struct dpif *dpif) { - poll_fd_wait(dpif->fd, POLLIN); + dpif->class->recv_wait(dpif); } +/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type' + * and '*engine_id', respectively. */ void dpif_get_netflow_ids(const struct dpif *dpif, uint8_t *engine_type, uint8_t *engine_id) { - *engine_type = *engine_id = dpif->minor; + *engine_type = dpif->netflow_engine_type; + *engine_id = dpif->netflow_engine_id; +} + +void +dpif_init(struct dpif *dpif, const struct dpif_class *class, const char *name, + uint8_t netflow_engine_type, uint8_t netflow_engine_id) +{ + dpif->class = class; + dpif->name = xstrdup(name); + dpif->netflow_engine_type = netflow_engine_type; + dpif->netflow_engine_id = netflow_engine_id; +} + +static void +log_operation(const struct dpif *dpif, const char *operation, int error) +{ + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation); + } else { + VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", + dpif_name(dpif), operation, strerror(error)); + } +} + +static enum vlog_level +flow_message_log_level(int error) +{ + return error ? VLL_WARN : VLL_DBG; +} + +static bool +should_log_flow_message(int error) +{ + return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error), + error ? &error_rl : &dpmsg_rl); +} + +static void +log_flow_message(const struct dpif *dpif, int error, const char *operation, + const flow_t *flow, const struct odp_flow_stats *stats, + const union odp_action *actions, size_t n_actions) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + ds_put_format(&ds, "%s: ", dpif_name(dpif)); + if (error) { + ds_put_cstr(&ds, "failed to "); + } + ds_put_format(&ds, "%s ", operation); + if (error) { + ds_put_format(&ds, "(%s) ", strerror(error)); + } + flow_format(&ds, flow); + if (stats) { + ds_put_cstr(&ds, ", "); + format_odp_flow_stats(&ds, stats); + } + if (actions || n_actions) { + ds_put_cstr(&ds, ", actions:"); + format_odp_actions(&ds, actions, n_actions); + } + vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +log_flow_operation(const struct dpif *dpif, const char *operation, int error, + struct odp_flow *flow) +{ + if (error) { + flow->n_actions = 0; + } + log_flow_message(dpif, error, operation, &flow->key, + !error ? &flow->stats : NULL, + flow->actions, flow->n_actions); +} + +static void +log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put) +{ + enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS }; + struct ds s; + + ds_init(&s); + ds_put_cstr(&s, "put"); + if (put->flags & ODPPF_CREATE) { + ds_put_cstr(&s, "[create]"); + } + if (put->flags & ODPPF_MODIFY) { + ds_put_cstr(&s, "[modify]"); + } + if (put->flags & ODPPF_ZERO_STATS) { + ds_put_cstr(&s, "[zero]"); + } + if (put->flags & ~ODPPF_ALL) { + ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL); + } + log_flow_message(dpif, error, ds_cstr(&s), &put->flow.key, + !error ? &put->flow.stats : NULL, + put->flow.actions, put->flow.n_actions); + ds_destroy(&s); +} + +/* There is a tendency to construct odp_flow objects on the stack and to + * forget to properly initialize their "actions" and "n_actions" members. + * When this happens, we get memory corruption because the kernel + * writes through the random pointer that is in the "actions" member. + * + * This function attempts to combat the problem by: + * + * - Forcing a segfault if "actions" points to an invalid region (instead + * of just getting back EFAULT, which can be easily missed in the log). + * + * - Storing a distinctive value that is likely to cause an + * easy-to-identify error later if it is dereferenced, etc. + * + * - Triggering a warning on uninitialized memory from Valgrind if + * "actions" or "n_actions" was not initialized. + */ +static void +check_rw_odp_flow(struct odp_flow *flow) +{ + if (flow->n_actions) { + memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]); + } } +#include +#include +#include +#include +#include +#include +#include +#include +#include + struct dpifmon { struct dpif *dpif; struct nl_sock *sock; @@ -895,247 +1042,3 @@ dpifmon_wait(struct dpifmon *mon) { nl_sock_wait(mon->sock, POLLIN); } - -static int get_openvswitch_major(void); -static int get_major(const char *target, int default_major); - -static int -lookup_minor(const char *name, unsigned int *minor) -{ - struct ethtool_drvinfo drvinfo; - struct ifreq ifr; - int error; - int sock; - - *minor = -1; - sock = socket(AF_INET, SOCK_DGRAM, 0); - if (sock < 0) { - VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno)); - error = errno; - goto error; - } - - memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - ifr.ifr_data = (caddr_t) &drvinfo; - - memset(&drvinfo, 0, sizeof drvinfo); - drvinfo.cmd = ETHTOOL_GDRVINFO; - if (ioctl(sock, SIOCETHTOOL, &ifr)) { - VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno)); - error = errno; - goto error_close_sock; - } - - if (strcmp(drvinfo.driver, "openvswitch")) { - VLOG_WARN("%s is not an openvswitch device", name); - error = EOPNOTSUPP; - goto error_close_sock; - } - - if (!isdigit(drvinfo.bus_info[0])) { - VLOG_WARN("%s ethtool info does not contain an openvswitch minor", - name); - error = EPROTOTYPE; - goto error_close_sock; - } - - *minor = atoi(drvinfo.bus_info); - close(sock); - return 0; - -error_close_sock: - close(sock); -error: - return error; -} - -static int -make_openvswitch_device(unsigned int minor, char **fnp) -{ - dev_t dev = makedev(get_openvswitch_major(), minor); - const char dirname[] = "/dev/net"; - struct stat s; - char fn[128]; - - *fnp = NULL; - sprintf(fn, "%s/dp%d", dirname, minor); - if (!stat(fn, &s)) { - if (!S_ISCHR(s.st_mode)) { - VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing", - fn); - } else if (s.st_rdev != dev) { - VLOG_WARN_RL(&error_rl, - "%s is device %u:%u instead of %u:%u, fixing", - fn, major(s.st_rdev), minor(s.st_rdev), - major(dev), minor(dev)); - } else { - goto success; - } - if (unlink(fn)) { - VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)", - fn, strerror(errno)); - return errno; - } - } else if (errno == ENOENT) { - if (stat(dirname, &s)) { - if (errno == ENOENT) { - if (mkdir(dirname, 0755)) { - VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)", - dirname, strerror(errno)); - return errno; - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", - dirname, strerror(errno)); - return errno; - } - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno)); - return errno; - } - - /* The device needs to be created. */ - if (mknod(fn, S_IFCHR | 0700, dev)) { - VLOG_WARN_RL(&error_rl, - "%s: creating character device %u:%u failed (%s)", - fn, major(dev), minor(dev), strerror(errno)); - return errno; - } - -success: - *fnp = xstrdup(fn); - return 0; -} - - -static int -get_openvswitch_major(void) -{ - static unsigned int openvswitch_major; - if (!openvswitch_major) { - enum { DEFAULT_MAJOR = 248 }; - openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR); - } - return openvswitch_major; -} - -static int -get_major(const char *target, int default_major) -{ - const char fn[] = "/proc/devices"; - char line[128]; - FILE *file; - int ln; - - file = fopen(fn, "r"); - if (!file) { - VLOG_ERR("opening %s failed (%s)", fn, strerror(errno)); - goto error; - } - - for (ln = 1; fgets(line, sizeof line, file); ln++) { - char name[64]; - int major; - - if (!strncmp(line, "Character", 9) || line[0] == '\0') { - /* Nothing to do. */ - } else if (!strncmp(line, "Block", 5)) { - /* We only want character devices, so skip the rest of the file. */ - break; - } else if (sscanf(line, "%d %63s", &major, name)) { - if (!strcmp(name, target)) { - fclose(file); - return major; - } - } else { - static bool warned; - if (!warned) { - VLOG_WARN("%s:%d: syntax error", fn, ln); - } - warned = true; - } - } - - VLOG_ERR("%s: %s major not found (is the module loaded?), using " - "default major %d", fn, target, default_major); -error: - VLOG_INFO("using default major %d for %s", default_major, target); - return default_major; -} - -static int -name_to_minor(const char *name, unsigned int *minor) -{ - if (!get_minor_from_name(name, minor)) { - return 0; - } - return lookup_minor(name, minor); -} - -static int -get_minor_from_name(const char *name, unsigned int *minor) -{ - if (!strncmp(name, "dp", 2) && isdigit(name[2])) { - *minor = atoi(name + 2); - return 0; - } else { - return EINVAL; - } -} - -static int -open_by_minor(unsigned int minor, struct dpif **dpifp) -{ - struct dpif *dpif; - int error; - char *fn; - int fd; - - *dpifp = NULL; - error = make_openvswitch_device(minor, &fn); - if (error) { - return error; - } - - fd = open(fn, O_RDONLY | O_NONBLOCK); - if (fd < 0) { - error = errno; - VLOG_WARN("%s: open failed (%s)", fn, strerror(error)); - free(fn); - return error; - } - free(fn); - - dpif = xmalloc(sizeof *dpif); - dpif->name = xasprintf("dp%u", dpif->minor); - dpif->minor = minor; - dpif->fd = fd; - *dpifp = dpif; - return 0; -} - -/* There is a tendency to construct odp_flow objects on the stack and to - * forget to properly initialize their "actions" and "n_actions" members. - * When this happens, we get memory corruption because the kernel - * writes through the random pointer that is in the "actions" member. - * - * This function attempts to combat the problem by: - * - * - Forcing a segfault if "actions" points to an invalid region (instead - * of just getting back EFAULT, which can be easily missed in the log). - * - * - Storing a distinctive value that is likely to cause an - * easy-to-identify error later if it is dereferenced, etc. - * - * - Triggering a warning on uninitialized memory from Valgrind if - * "actions" or "n_actions" was not initialized. - */ -static void -check_rw_odp_flow(struct odp_flow *flow) -{ - if (flow->n_actions) { - memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]); - } -} diff --git a/lib/dpif.h b/lib/dpif.h index 88bd15dc..5302197e 100644 --- a/lib/dpif.h +++ b/lib/dpif.h @@ -53,10 +53,10 @@ int dpif_port_get_name(struct dpif *, uint16_t port_no, char *name, size_t name_size); int dpif_port_list(const struct dpif *, struct odp_port **, size_t *n_ports); -int dpif_port_group_set(struct dpif *, uint16_t group, - const uint16_t ports[], size_t n_ports); int dpif_port_group_get(const struct dpif *, uint16_t group, uint16_t **ports, size_t *n_ports); +int dpif_port_group_set(struct dpif *, uint16_t group, + const uint16_t ports[], size_t n_ports); int dpif_flow_flush(struct dpif *); int dpif_flow_put(struct dpif *, struct odp_flow_put *); diff --git a/lib/vlog-modules.def b/lib/vlog-modules.def index 216111b1..2653781b 100644 --- a/lib/vlog-modules.def +++ b/lib/vlog-modules.def @@ -30,6 +30,7 @@ VLOG_MODULE(dhcp) VLOG_MODULE(dhcp_client) VLOG_MODULE(discovery) VLOG_MODULE(dpif) +VLOG_MODULE(dpif_linux) VLOG_MODULE(dpctl) VLOG_MODULE(executer) VLOG_MODULE(ezio_term)